Datasets:

james-burton

/

textual-explanations-702010

like 0

GradientBoostingClassifier

C2

Food Ordering Customer Churn Prediction

The prediction probability of C1 is 17.93% and that of C2 is 82.07%. Therefore, the most probable class for the given case is C2. The above classification assertion statements are based on the information supplied to the classifier about the case given. The top features with significant attributions leading to the decision made above are F6, F26, F32, F42, F43, and F25. Conversely, F38, F24, F14, F28, and F33 are among the features deemed irrelevant to the classification decision here since their contributions are almost negligible and much closer to zero. The attribution analysis suggests that not all the relevant features positively contribute to the classifier's arriving at the verdict here. Those with positive attributions that push the classifier towards generating C2 as the label are F6, F26, F32, F42, F4, F7, F16, and F21. Decreasing the likelihood of the correctness of C2 are the negative features such as F43, F29, F25, F13, F39, F5, F41, and F12, which could be blamed for the little uncertainty in the classification output, as indicated by the prediction probability of C1.

{'C1': '17.93%', 'C2': '82.07%'}

{'F6': 'More restaurant choices', 'F26': 'Ease and convenient', 'F32': 'Bad past experience', 'F42': 'Time saving', 'F25': 'Unaffordable', 'F43': 'Educational Qualifications', 'F29': 'Late Delivery', 'F7': 'Occupation', 'F39': 'Influence of rating', 'F5': 'Less Delivery time', 'F13': 'Order placed by mistake', 'F41': 'Delivery person ability', 'F12': 'Order Time', 'F4': 'Unavailability', 'F20': 'More Offers and Discount', 'F35': 'Delay of delivery person picking up food', 'F16': 'Good Taste ', 'F21': 'Wrong order delivered', 'F10': 'Freshness ', 'F1': 'Missing item', 'F38': 'Residence in busy location', 'F14': 'Google Maps Accuracy', 'F24': 'Age', 'F33': 'Good Road Condition', 'F28': 'Low quantity low time', 'F19': 'High Quality of package', 'F22': 'Number of calls', 'F3': 'Politeness', 'F27': 'Temperature', 'F17': 'Maximum wait time', 'F8': 'Long delivery time', 'F18': 'Influence of time', 'F15': 'Delay of delivery person getting assigned', 'F40': 'Family size', 'F31': 'Poor Hygiene', 'F30': 'Health Concern', 'F23': 'Self Cooking', 'F44': 'Good Tracking system', 'F37': 'Good Food quality', 'F9': 'Easy Payment option', 'F45': 'Perference(P2)', 'F46': 'Perference(P1)', 'F34': 'Monthly Income', 'F2': 'Marital Status', 'F36': 'Gender', 'F11': 'Good Quantity'}

{'F12': 'F6', 'F10': 'F26', 'F21': 'F32', 'F11': 'F42', 'F23': 'F25', 'F6': 'F43', 'F19': 'F29', 'F4': 'F7', 'F38': 'F39', 'F39': 'F5', 'F29': 'F13', 'F37': 'F41', 'F31': 'F12', 'F22': 'F4', 'F14': 'F20', 'F26': 'F35', 'F45': 'F16', 'F27': 'F21', 'F43': 'F10', 'F28': 'F1', 'F33': 'F38', 'F34': 'F14', 'F1': 'F24', 'F35': 'F33', 'F36': 'F28', 'F40': 'F19', 'F41': 'F22', 'F42': 'F3', 'F44': 'F27', 'F32': 'F17', 'F24': 'F8', 'F30': 'F18', 'F25': 'F15', 'F7': 'F40', 'F20': 'F31', 'F18': 'F30', 'F17': 'F23', 'F16': 'F44', 'F15': 'F37', 'F13': 'F9', 'F9': 'F45', 'F8': 'F46', 'F5': 'F34', 'F3': 'F2', 'F2': 'F36', 'F46': 'F11'}

{'C1': 'C1', 'C2': 'C2'}

Go Away

{'C1': 'Return', 'C2': 'Go Away'}

DNN

C1

Credit Card Fraud Classification

The classification algorithm classifies the given case as C1 with a confidence level equal to 99.99%, suggesting that there is little chance that the C2 label could be the true label. The classification confidence level can be attributed to the influence and contributions of the features F29, F8, F10, F26, and F23. Positively supporting the model's decision are values of F29, F8, F10, and F26. On the contrary, the values of F23, F4, F3, and F7 are shifting the model towards producing the C2 label, which results in a marginal decrease in the certainty associated with the C1 label. The other positively supported features further improving the odds in favour of C1 include F25, F12, F16, and F5. Overall, it is not farfetched to accept that C1 is the correct label for the case under consideration since the strong positive influences of F29, F8, and F10 far outweigh the influence of any of the other input features. In other words, as mentioned above, there is only a small chance that the true label is not C1 considering the attributions of the top influential input features.

{'C1': '99.99%', 'C2': '0.01%'}

{'F29': 'Z3', 'F8': 'Z6', 'F10': 'Time', 'F23': 'Z13', 'F26': 'Z12', 'F4': 'Z4', 'F7': 'Z10', 'F25': 'Z5', 'F3': 'Z9', 'F24': 'Z14', 'F27': 'Z16', 'F30': 'Z11', 'F12': 'Z17', 'F21': 'Z19', 'F16': 'Z8', 'F9': 'Z28', 'F28': 'Z21', 'F20': 'Z20', 'F13': 'Z1', 'F22': 'Z24', 'F6': 'Z18', 'F5': 'Z2', 'F14': 'Z25', 'F1': 'Amount', 'F2': 'Z26', 'F17': 'Z27', 'F15': 'Z22', 'F19': 'Z15', 'F11': 'Z7', 'F18': 'Z23'}

{'F4': 'F29', 'F7': 'F8', 'F1': 'F10', 'F14': 'F23', 'F13': 'F26', 'F5': 'F4', 'F11': 'F7', 'F6': 'F25', 'F10': 'F3', 'F15': 'F24', 'F17': 'F27', 'F12': 'F30', 'F18': 'F12', 'F20': 'F21', 'F9': 'F16', 'F29': 'F9', 'F22': 'F28', 'F21': 'F20', 'F2': 'F13', 'F25': 'F22', 'F19': 'F6', 'F3': 'F5', 'F26': 'F14', 'F30': 'F1', 'F27': 'F2', 'F28': 'F17', 'F23': 'F15', 'F16': 'F19', 'F8': 'F11', 'F24': 'F18'}

{'C1': 'C1', 'C2': 'C2'}

Not Fraud

{'C1': 'Not Fraud', 'C2': 'Fraud'}

RandomForestClassifier

C2

Employee Attrition

The data is marked as C2 by the classifier based on the input features, with a moderate degree of confidence since the prediction probability of the other label, C1, is only 44.0%. The most influential features driving the classification above are F27, F12, F24, F14, F30, F18, F26, F2, F20, F22, F13, F21, F17, F23, F10, F25, F5, F6, and F8. Strongly reducing the chance of C2 being the true label for the given case are the negative features F12 and F27. Actually, these negative features, along with other features such as F30, F18, and F20, are responsible for the uncertainty in the classification decision here. On the contrary, the input features F24, F14, F26, F2, F22, and F13 positively contribute to the classifier's decision to choose C2 as the label here. Finally, it is important to note that not all the features are shown to be relevant when making the labelling decision regarding the case under consideration, and these irrelevant features include F15, F4, F1, and F29.

{'C1': '44.00%', 'C2': '56.00%'}

{'F27': 'OverTime', 'F12': 'BusinessTravel', 'F24': 'MaritalStatus', 'F14': 'JobInvolvement', 'F30': 'WorkLifeBalance', 'F18': 'Education', 'F26': 'EnvironmentSatisfaction', 'F2': 'Gender', 'F20': 'JobRole', 'F22': 'NumCompaniesWorked', 'F13': 'YearsInCurrentRole', 'F21': 'HourlyRate', 'F17': 'Department', 'F23': 'RelationshipSatisfaction', 'F10': 'PerformanceRating', 'F28': 'YearsWithCurrManager', 'F25': 'Age', 'F5': 'MonthlyRate', 'F6': 'StockOptionLevel', 'F8': 'JobSatisfaction', 'F15': 'DailyRate', 'F7': 'YearsSinceLastPromotion', 'F9': 'YearsAtCompany', 'F29': 'TrainingTimesLastYear', 'F11': 'EducationField', 'F1': 'TotalWorkingYears', 'F16': 'PercentSalaryHike', 'F19': 'MonthlyIncome', 'F3': 'JobLevel', 'F4': 'DistanceFromHome'}

{'F26': 'F27', 'F17': 'F12', 'F25': 'F24', 'F29': 'F14', 'F20': 'F30', 'F27': 'F18', 'F28': 'F26', 'F23': 'F2', 'F24': 'F20', 'F8': 'F22', 'F14': 'F13', 'F4': 'F21', 'F21': 'F17', 'F18': 'F23', 'F19': 'F10', 'F16': 'F28', 'F1': 'F25', 'F7': 'F5', 'F10': 'F6', 'F30': 'F8', 'F2': 'F15', 'F15': 'F7', 'F13': 'F9', 'F12': 'F29', 'F22': 'F11', 'F11': 'F1', 'F9': 'F16', 'F6': 'F19', 'F5': 'F3', 'F3': 'F4'}

{'C1': 'C1', 'C2': 'C2'}

Leave

{'C1': 'Leave', 'C2': 'Leave'}

RandomForestClassifier

C1

Cab Surge Pricing System

The model determined that this case belongs to C1 of the three possible labels, with an 83.0% likelihood. It is important to note, however, that there is about a 14.0% chance that it could be C2 and a 3.0% chance that it is rather C3. The most relevant feature driving this prediction is F12, with a very strong positive attribution, increasing the odds of the label C1. The following attributes have values pushing for a different prediction: F11, F4, F7, and F6, however, their attributions are very low when compared to that from F12. Other features positively contributing to the model's decision for this test case are F5, F2, F9, F1, F3, F10, and F8, with F3, F10, and F8 being the least relevant features considered by the model for the given case.

{'C3': '3.00%', 'C1': '83.00%', 'C2': '14.00%'}

{'F12': 'Type_of_Cab', 'F11': 'Destination_Type', 'F4': 'Trip_Distance', 'F7': 'Cancellation_Last_1Month', 'F5': 'Confidence_Life_Style_Index', 'F2': 'Var3', 'F6': 'Customer_Since_Months', 'F9': 'Life_Style_Index', 'F1': 'Var2', 'F3': 'Gender', 'F10': 'Var1', 'F8': 'Customer_Rating'}

{'F2': 'F12', 'F6': 'F11', 'F1': 'F4', 'F8': 'F7', 'F5': 'F5', 'F11': 'F2', 'F3': 'F6', 'F4': 'F9', 'F10': 'F1', 'F12': 'F3', 'F9': 'F10', 'F7': 'F8'}

{'C1': 'C3', 'C2': 'C1', 'C3': 'C2'}

C2

{'C3': 'Low', 'C1': 'Medium', 'C2': 'High'}

RandomForestClassifier

C1

Annual Income Earnings

The classifier assigned the label C1, given that there is merely a 2.18% chance that C2 is the correct label. Influencing this classification decision are mainly the values of the variables F5, F12, F11, and F4 which are also commonly referred to as positive variables since they increase the response in favour of the predicted label. Other variables supporting the prediction of C1 are F1, F14, F6, and F8. However, unlike F5, F12, F11, and F4, these variables have a moderate impact on the classifier. The variables that decrease the likelihood that C1 is the correct label are F3, F2, F13, and F7 since they have values that swing the classification verdict in the direction of C2.

{'C1': '97.82%', 'C2': '2.18%'}

{'F5': 'Capital Gain', 'F12': 'Marital Status', 'F11': 'Relationship', 'F4': 'Age', 'F1': 'Education-Num', 'F14': 'Hours per week', 'F3': 'Occupation', 'F6': 'Capital Loss', 'F2': 'Sex', 'F8': 'Education', 'F13': 'Race', 'F9': 'fnlwgt', 'F7': 'Country', 'F10': 'Workclass'}

{'F11': 'F5', 'F6': 'F12', 'F8': 'F11', 'F1': 'F4', 'F5': 'F1', 'F13': 'F14', 'F7': 'F3', 'F12': 'F6', 'F10': 'F2', 'F4': 'F8', 'F9': 'F13', 'F3': 'F9', 'F14': 'F7', 'F2': 'F10'}

{'C1': 'C1', 'C2': 'C2'}

Under 50K

{'C1': 'Under 50K', 'C2': 'Above 50K'}

SVMClassifier_poly

C1

Employee Attrition

Because the chance that C2 is the right label is around 42.17 percent, the example under review is labelled as C1 with a moderate degree of confidence. F4, F1, F7, F15, F6, and F14 have the most influence on the above forecast, whereas F25, F21, F11, F13, F29, F10, and F26 have small contributions. F20, F22, F5, F8, F17, F12, and F24 all have a relatively modest impact. However, the classifier does not take into account all of the attributes while making a judgement in a specific case and the attributes F9, F2, F16, and F23 are all irrelevant features. F4, F1, F7, F6, F21, F29, and F12 are the positive features pushing the prediction in support of the forecasted label. We can see from the attributions map that the bulk of the influential features exhibit negative attributions that reduce the likelihood that C1 is the correct label, justifying the uncertainty associated with the classifier's prediction choice.

{'C1': '57.83%', 'C2': '42.17%'}

{'F4': 'OverTime', 'F1': 'NumCompaniesWorked', 'F7': 'RelationshipSatisfaction', 'F15': 'MaritalStatus', 'F6': 'YearsSinceLastPromotion', 'F14': 'Department', 'F25': 'Age', 'F21': 'Education', 'F11': 'EducationField', 'F13': 'BusinessTravel', 'F29': 'JobLevel', 'F10': 'JobInvolvement', 'F26': 'WorkLifeBalance', 'F20': 'MonthlyRate', 'F22': 'YearsAtCompany', 'F5': 'Gender', 'F8': 'PerformanceRating', 'F17': 'JobRole', 'F12': 'TrainingTimesLastYear', 'F24': 'EnvironmentSatisfaction', 'F9': 'YearsWithCurrManager', 'F2': 'DailyRate', 'F16': 'YearsInCurrentRole', 'F23': 'TotalWorkingYears', 'F27': 'StockOptionLevel', 'F3': 'PercentSalaryHike', 'F28': 'MonthlyIncome', 'F30': 'HourlyRate', 'F18': 'DistanceFromHome', 'F19': 'JobSatisfaction'}

{'F26': 'F4', 'F8': 'F1', 'F18': 'F7', 'F25': 'F15', 'F15': 'F6', 'F21': 'F14', 'F1': 'F25', 'F27': 'F21', 'F22': 'F11', 'F17': 'F13', 'F5': 'F29', 'F29': 'F10', 'F20': 'F26', 'F7': 'F20', 'F13': 'F22', 'F23': 'F5', 'F19': 'F8', 'F24': 'F17', 'F12': 'F12', 'F28': 'F24', 'F16': 'F9', 'F2': 'F2', 'F14': 'F16', 'F11': 'F23', 'F10': 'F27', 'F9': 'F3', 'F6': 'F28', 'F4': 'F30', 'F3': 'F18', 'F30': 'F19'}

{'C1': 'C1', 'C2': 'C2'}

Leave

{'C1': 'Leave', 'C2': 'Leave'}

BernoulliNB

C1

Credit Card Fraud Classification

All features are shown to have a positive impact on the classification to class C1 or to have no impact at all. F12, F1, F17, and F28 are the four features with the most impact. Some of the remaining features, in order of feature importance, are F5, F7, F10, F27, F21, F18, F2, F23, F24, F29, F11, and F6. F12 and F1 both have the highest positive impact on the final classification, pushing the classification towards class C1. All of F17, F28, F5, and F7 influence the model's classification to C1. In terms of the features which have a positive impact on the classification, features F10, F27, F21, and F18 are all ranked to have a medium degree of influence on the final classification. F10 and F27 both have a similar importance attribution, which is higher than that of F21 and F18. All the other features not listed above are irrelevant to the decision above and among them are F22, F25, and F30.

{'C2': '5.16%', 'C1': '94.84%'}

{'F12': 'Z14', 'F1': 'Z1', 'F17': 'Z17', 'F28': 'Amount', 'F5': 'Z19', 'F7': 'Z5', 'F10': 'Z3', 'F27': 'Z8', 'F21': 'Z18', 'F18': 'Z10', 'F2': 'Z26', 'F23': 'Z25', 'F24': 'Z22', 'F29': 'Z4', 'F11': 'Z7', 'F6': 'Z13', 'F26': 'Z23', 'F19': 'Z9', 'F8': 'Z21', 'F4': 'Z2', 'F22': 'Z28', 'F30': 'Z24', 'F25': 'Z27', 'F14': 'Time', 'F16': 'Z20', 'F9': 'Z16', 'F20': 'Z12', 'F3': 'Z11', 'F15': 'Z6', 'F13': 'Z15'}

{'F15': 'F12', 'F2': 'F1', 'F18': 'F17', 'F30': 'F28', 'F20': 'F5', 'F6': 'F7', 'F4': 'F10', 'F9': 'F27', 'F19': 'F21', 'F11': 'F18', 'F27': 'F2', 'F26': 'F23', 'F23': 'F24', 'F5': 'F29', 'F8': 'F11', 'F14': 'F6', 'F24': 'F26', 'F10': 'F19', 'F22': 'F8', 'F3': 'F4', 'F29': 'F22', 'F25': 'F30', 'F28': 'F25', 'F1': 'F14', 'F21': 'F16', 'F17': 'F9', 'F13': 'F20', 'F12': 'F3', 'F7': 'F15', 'F16': 'F13'}

{'C1': 'C2', 'C2': 'C1'}

Fraud

{'C2': 'Not Fraud', 'C1': 'Fraud'}

LogisticRegression

C1

Used Cars Price-Range Prediction

Label C2 has a lower probability than label C1, so C1 is the most likely option in this case. C1 has a probability of approximately 96.25 percent, which can be attributed to variables such as F7, F2, F8, and F6. According to the attributions assessment, the least relevant variables are F3, F4, and F9. Inspection of the direction of influence of the features showed that F10 and F1 present negative contributions that push the model somewhat away from producing C1 because they support the label C2. Considering that the combined impact of the negative variables is quite minimal in comparison to the combined impact of the positive variables such as F7, F2, F8, F5, and F6, it is not surprising that the model is very certain that C2 is not the accurate label for the given case.

{'C1': '96.25%', 'C2': '3.75%'}

{'F2': 'Fuel_Type', 'F7': 'Power', 'F8': 'Engine', 'F6': 'Seats', 'F10': 'car_age', 'F1': 'Owner_Type', 'F5': 'Name', 'F3': 'Mileage', 'F4': 'Kilometers_Driven', 'F9': 'Transmission'}

{'F7': 'F2', 'F4': 'F7', 'F3': 'F8', 'F10': 'F6', 'F5': 'F10', 'F9': 'F1', 'F6': 'F5', 'F2': 'F3', 'F1': 'F4', 'F8': 'F9'}

{'C1': 'C1', 'C2': 'C2'}

Low

{'C1': 'Low', 'C2': 'High'}

GradientBoostingClassifier

C2

Paris House Classification

According to the prediction made here, the most likely label for the given case is C2, with a prediction probability of 97.02%, indicating that the prediction probability of C1 is only 2.98%. The classification above is mainly due to the influence of F9, F13, and F5. The next set of features with moderate contributions includes F1, F12, and F6. However, those with little consideration from the classifier are F7, F4, F10, and F15. In consideration of the fact that all the top four features have a strong positive contribution, it is foreseeable why the classifier is relatively confident that the correct label for this case is C2. Additionally, the negative attributes with moderate to low impact are F12, F3, and F16.

{'C1': '2.98%', 'C2': '97.02%'}

{'F9': 'isNewBuilt', 'F13': 'hasYard', 'F5': 'hasPool', 'F1': 'hasStormProtector', 'F12': 'made', 'F6': 'squareMeters', 'F3': 'floors', 'F16': 'cityCode', 'F14': 'hasGuestRoom', 'F17': 'basement', 'F2': 'numPrevOwners', 'F11': 'price', 'F8': 'numberOfRooms', 'F7': 'garage', 'F4': 'cityPartRange', 'F10': 'hasStorageRoom', 'F15': 'attic'}

{'F3': 'F9', 'F1': 'F13', 'F2': 'F5', 'F4': 'F1', 'F12': 'F12', 'F6': 'F6', 'F8': 'F3', 'F9': 'F16', 'F16': 'F14', 'F13': 'F17', 'F11': 'F2', 'F17': 'F11', 'F7': 'F8', 'F15': 'F7', 'F10': 'F4', 'F5': 'F10', 'F14': 'F15'}

{'C1': 'C1', 'C2': 'C2'}

Luxury

{'C1': 'Basic', 'C2': 'Luxury'}

RandomForestClassifier

C1

Flight Price-Range Classification

The classification conclusion is as follows: C1 is the most likely label for this case and the classifier is certain that neither C3 nor C2 are the right labels since their likelihoods are equal to zero. The driving factors for the above classification are F6, F11, and F12, all of which have a substantial positive impact, causing the classifier to select C1. F1, F4, F2, and F9 are also positive features. The assigned label is not supported by all of the input features since the negative features F10, F8, and F3 support the decision that the most likely class for this instance could be any one of the other labels, C2 and C3. Nevertheless, given the confidence level in the aforementioned classification, it is reasonable to assume that the classifier paid little attention to the negative features, resulting in the selection of class C1.

{'C1': '100.00%', 'C3': '0.00%', 'C2': '0.00%'}

{'F6': 'Airline', 'F11': 'Duration_hours', 'F12': 'Total_Stops', 'F1': 'Journey_month', 'F10': 'Source', 'F4': 'Destination', 'F2': 'Arrival_hour', 'F8': 'Journey_day', 'F3': 'Dep_minute', 'F9': 'Arrival_minute', 'F5': 'Duration_mins', 'F7': 'Dep_hour'}

{'F9': 'F6', 'F7': 'F11', 'F12': 'F12', 'F2': 'F1', 'F10': 'F10', 'F11': 'F4', 'F5': 'F2', 'F1': 'F8', 'F4': 'F3', 'F6': 'F9', 'F8': 'F5', 'F3': 'F7'}

{'C1': 'C1', 'C2': 'C3', 'C3': 'C2'}

Low

{'C1': 'Low', 'C3': 'Moderate', 'C2': 'High'}

KNeighborsClassifier

C2

Water Quality Classification

The classifier states that there is a 50.0% chance that the true label of this test observation is C2. This indicates that the classifier is less certain in its prediction decision regarding the case under consideration. The label assigned is mainly due to the values of the features F9, F6, F4, F1, F2, and F7. The top features, F9 and F6, have very strong positive contributions pushing the prediction higher towards the most probable label. Among the remaining features stated above, F4, F1, F2, and F7, only F7 demonstrates some level of contradiction, forcing the labelling decision in a different direction. Finally, the features with marginal impact on the prediction made here are F5, F8, and F3. While F5 and F3 positively influence the decision made, F8 suggests that the label assigned by the classifier might not be the true label.

{'C2': '50.00%', 'C1': '50.00%'}

{'F9': 'Hardness', 'F6': 'Sulfate', 'F4': 'Organic_carbon', 'F1': 'Solids', 'F2': 'Conductivity', 'F7': 'Trihalomethanes', 'F5': 'ph', 'F8': 'Turbidity', 'F3': 'Chloramines'}

{'F2': 'F9', 'F5': 'F6', 'F7': 'F4', 'F3': 'F1', 'F6': 'F2', 'F8': 'F7', 'F1': 'F5', 'F9': 'F8', 'F4': 'F3'}

{'C1': 'C2', 'C2': 'C1'}

Not Portable

{'C2': 'Not Portable', 'C1': 'Portable'}

GradientBoostingClassifier

C2

Printer Sales

According to the attribution analysis, the each input variables contributes differently to the decision. For the case under consideration, there are variables that have negative influence on the decision here, but it also has numerous quantifiable variables that are positive. Per the model, C2 is 91.95% certain to be the correct label and C1 has a predicted probability of only 8.05%. The most essential input variables are F1, F7, F8, and F2, which allow the model to effectively compute the likelihoods across the classes, C2 and C1. F26 and F17 have nearly comparable positive effects, but F15 and F3 have a negative influence, altering the output decision in favour of a different label. The cumulative positive contribution of F26, F1, F2, F4, F19, and F17 was greater than that of F7, F8, F3, and F15, hence the positive variables succeed at improving the predictability odds in favour of the C2 class. Furthermore per the variable attributions, the contributions of F22, F5, F20, and F9 has very little to do with the classification decision since their attributions are negligible and closer to zero than all the above-mentioned variables.

{'C1': '8.05%', 'C2': '91.95%'}

{'F1': 'X24', 'F7': 'X8', 'F2': 'X1', 'F8': 'X21', 'F3': 'X4', 'F17': 'X6', 'F26': 'X3', 'F15': 'X22', 'F4': 'X7', 'F19': 'X15', 'F6': 'X20', 'F24': 'X11', 'F25': 'X10', 'F13': 'X19', 'F10': 'X5', 'F23': 'X16', 'F21': 'X23', 'F12': 'X9', 'F11': 'X17', 'F16': 'X18', 'F14': 'X25', 'F18': 'X14', 'F9': 'X2', 'F20': 'X13', 'F5': 'X12', 'F22': 'X26'}

{'F24': 'F1', 'F8': 'F7', 'F1': 'F2', 'F21': 'F8', 'F4': 'F3', 'F6': 'F17', 'F3': 'F26', 'F22': 'F15', 'F7': 'F4', 'F15': 'F19', 'F20': 'F6', 'F11': 'F24', 'F10': 'F25', 'F19': 'F13', 'F5': 'F10', 'F16': 'F23', 'F23': 'F21', 'F9': 'F12', 'F17': 'F11', 'F18': 'F16', 'F25': 'F14', 'F14': 'F18', 'F2': 'F9', 'F13': 'F20', 'F12': 'F5', 'F26': 'F22'}

{'C1': 'C1', 'C2': 'C2'}

More

{'C1': 'Less', 'C2': 'More'}

LogisticRegression

C2

Annual Income Earnings

Deciding the most probable label for the given case on the basis of the values of the input variables, the classification algorithm's output decision is that: the probability of C2 being the correct label is 79.78%, the probability of C1 is 20.22%. Therefore, the most likely label is identified as C2 and the attribution analysis shows that all the variables contributed to some extent to the final decision by the algorithm with respect to the given case. The most influential variables are F9, F8, F14, and F7, but F13, F6, and F11 are the least influential ones. The analysis also indicates that F9, F7, F13, and F11 are responsible for the marginal doubt in the classification decision here hence they are commonly referred to as negative variables since their contributions only tend to shift the verdict in a different direction than the assigned label. Finally, the variables such as F8, F14, F4, F12, F2, and F1 are the positive variables that increase the algorithm's response in favour of outputting the C2 label.

{'C1': '20.22%', 'C2': '79.78%'}

{'F9': 'Capital Gain', 'F8': 'Marital Status', 'F14': 'Education', 'F7': 'Capital Loss', 'F4': 'Hours per week', 'F12': 'Sex', 'F2': 'Country', 'F1': 'Education-Num', 'F10': 'Occupation', 'F3': 'Race', 'F5': 'Age', 'F13': 'Workclass', 'F6': 'fnlwgt', 'F11': 'Relationship'}

{'F11': 'F9', 'F6': 'F8', 'F4': 'F14', 'F12': 'F7', 'F13': 'F4', 'F10': 'F12', 'F14': 'F2', 'F5': 'F1', 'F7': 'F10', 'F9': 'F3', 'F1': 'F5', 'F2': 'F13', 'F3': 'F6', 'F8': 'F11'}

{'C1': 'C1', 'C2': 'C2'}

Above 50K

{'C1': 'Under 50K', 'C2': 'Above 50K'}

LogisticRegression

C2

E-Commerce Shipping

The reliability of the classification verdict for this case is 71.57%, implying there is a 28.43% chance that the correct label could be C1. F6 has a significant negative impact on classification output since its contribution contradicts the labelling of the case as C2, hence favours labelling the case as C1. The values F5, F8, F7, F1, F4, F2, and F3 have a positive effect on the results, but still contributes less than the effect of F6. The analysis shows that F6 has an overwhelming negative impact or influence on the predictive decisions of the model here. F8, F7, F1, and F4 have a positive effect on model predictions. Due to the power of the F6 function, all other functions have little effect on the results. In summary, the uncertainty of the predictions can be explained by the control on the model by feature F6, which drags the classification decision favourably towards C1.

{'C2': '71.57%', 'C1': '28.43%'}

{'F6': 'Discount_offered', 'F5': 'Weight_in_gms', 'F8': 'Prior_purchases', 'F7': 'Product_importance', 'F1': 'Cost_of_the_Product', 'F4': 'Gender', 'F2': 'Customer_rating', 'F3': 'Warehouse_block', 'F10': 'Customer_care_calls', 'F9': 'Mode_of_Shipment'}

{'F2': 'F6', 'F3': 'F5', 'F8': 'F8', 'F9': 'F7', 'F1': 'F1', 'F10': 'F4', 'F7': 'F2', 'F4': 'F3', 'F6': 'F10', 'F5': 'F9'}

{'C1': 'C2', 'C2': 'C1'}

On-time

{'C2': 'On-time', 'C1': 'Late'}

DecisionTreeClassifier

C1

Vehicle Insurance Claims

This model predicted class label C1 with about 93.32% certainty, while there was about a 6.68% chance of the correct class being identified as a different label. Seven features, F30, F5, F33, F10, F11, F9, and F6, have higher impacts on the model prediction decision above. But the feature F30 has the largest positive impact on the result and on the contrary, F33, F11, F9, and F6 show the potential to shift the narrative to a different label since their contributions reduce the likelihood of the predicted label for this case. In addition, features F25, F12, and F15 have moderate impacts on the model's prediction but each of them is increasing the responses, and finally, the features shown have negligible influence include F17, and F3.

{'C1': '93.32%', 'C2': '6.68%'}

{'F30': 'incident_severity', 'F5': 'incident_city', 'F33': 'injury_claim', 'F10': 'insured_occupation', 'F11': 'insured_zip', 'F9': 'authorities_contacted', 'F6': 'auto_year', 'F25': 'police_report_available', 'F12': 'bodily_injuries', 'F15': 'insured_hobbies', 'F27': 'insured_sex', 'F1': 'auto_make', 'F24': 'property_damage', 'F31': 'witnesses', 'F22': 'insured_relationship', 'F7': 'age', 'F32': 'vehicle_claim', 'F4': 'months_as_customer', 'F20': 'property_claim', 'F18': 'incident_type', 'F23': 'capital-gains', 'F14': 'policy_deductable', 'F19': 'policy_annual_premium', 'F28': 'incident_state', 'F29': 'umbrella_limit', 'F26': 'total_claim_amount', 'F13': 'collision_type', 'F2': 'incident_hour_of_the_day', 'F16': 'insured_education_level', 'F17': 'number_of_vehicles_involved', 'F3': 'policy_csl', 'F8': 'policy_state', 'F21': 'capital-loss'}

{'F27': 'F30', 'F30': 'F5', 'F14': 'F33', 'F22': 'F10', 'F6': 'F11', 'F28': 'F9', 'F17': 'F6', 'F32': 'F25', 'F11': 'F12', 'F23': 'F15', 'F20': 'F27', 'F33': 'F1', 'F31': 'F24', 'F12': 'F31', 'F24': 'F22', 'F2': 'F7', 'F16': 'F32', 'F1': 'F4', 'F15': 'F20', 'F25': 'F18', 'F7': 'F23', 'F3': 'F14', 'F4': 'F19', 'F29': 'F28', 'F5': 'F29', 'F13': 'F26', 'F26': 'F13', 'F9': 'F2', 'F21': 'F16', 'F10': 'F17', 'F19': 'F3', 'F18': 'F8', 'F8': 'F21'}

{'C1': 'C1', 'C2': 'C2'}

Not Fraud

{'C1': 'Not Fraud', 'C2': 'Fraud'}

KNeighborsClassifier

C2

Printer Sales

Considering the prediction likelihoods, this case is labelled as C2 by the model, that is, the model states that there is about an 83.33% chance that the case is under C2 and about a 16.67% chance that it is not. The most relevant features influencing the decision made here are: F20, F7, F17, and F5. Among the feature set mentioned above, F20 and F7 offer a very strong positive contribution to the prediction of C2. Conversely, F5 suggests the alternative label C1 could be the true label for this case, but this attribution is weak when compared to F7 and F20. Other features that are moderately pushing for this classification decision include F8, F24, F25, and F16. However, the values of F21, F2, F11, and F26 advocate for the assignment of a different label. Finally, the features F6, F4, F14, and F22 have very little impact on the model's prediction for this case.

{'C2': '83.33%', 'C1': '16.67%'}

{'F20': 'X24', 'F7': 'X1', 'F5': 'X4', 'F17': 'X10', 'F8': 'X2', 'F24': 'X8', 'F21': 'X17', 'F25': 'X7', 'F16': 'X21', 'F2': 'X18', 'F11': 'X6', 'F26': 'X11', 'F10': 'X22', 'F15': 'X25', 'F18': 'X5', 'F1': 'X19', 'F19': 'X15', 'F9': 'X23', 'F12': 'X16', 'F23': 'X3', 'F6': 'X14', 'F4': 'X20', 'F14': 'X13', 'F22': 'X12', 'F13': 'X9', 'F3': 'X26'}

{'F24': 'F20', 'F1': 'F7', 'F4': 'F5', 'F10': 'F17', 'F2': 'F8', 'F8': 'F24', 'F17': 'F21', 'F7': 'F25', 'F21': 'F16', 'F18': 'F2', 'F6': 'F11', 'F11': 'F26', 'F22': 'F10', 'F25': 'F15', 'F5': 'F18', 'F19': 'F1', 'F15': 'F19', 'F23': 'F9', 'F16': 'F12', 'F3': 'F23', 'F14': 'F6', 'F20': 'F4', 'F13': 'F14', 'F12': 'F22', 'F9': 'F13', 'F26': 'F3'}

{'C1': 'C2', 'C2': 'C1'}

Less

{'C2': 'Less', 'C1': 'More'}

SVC

C1

Advertisement Prediction

For the given instance, the model generated the label C1 with a very high predicted probability equal to 99.66% which implies that the model is very confident that C2 is not the correct label. Ranking the contributions of the features to the prediction above, from the most relevant to the least relevant, is as follows: F2, F5, F1, F3, F4, F6, and F7. Among the seven features, only F4 and F6 have negative contributions, pushing the prediction towards the C2 label. However, given that these features have very low contributions, their impact on the model's decision is close to negligible when compared to the contributions of the positive features F2, F5, and F1.

{'C2': '0.34%', 'C1': '99.66%'}

{'F2': 'Daily Time Spent on Site', 'F5': 'Daily Internet Usage', 'F1': 'Age', 'F3': 'Gender', 'F4': 'ad_day', 'F6': 'ad_month', 'F7': 'Area Income'}

{'F1': 'F2', 'F4': 'F5', 'F2': 'F1', 'F5': 'F3', 'F7': 'F4', 'F6': 'F6', 'F3': 'F7'}

{'C1': 'C2', 'C2': 'C1'}

Watch

{'C2': 'Skip', 'C1': 'Watch'}

LogisticRegression

C1

Student Job Placement

The final prediction given by the model was C1 with almost 100% certainty, showing the model is confident about its decision. F10 had significantly more influence on the prediction than any other feature with F3 and F5 having the next highest attribution values. All the top features, F10, F3, and F5, encouraged the model to output class C1. F2, F12, and F9 are the features that had the least positive impact on the final classification. The features F4, F6, F8, and F11 have moderate impacts, pushing the model slightly away from a C1 classification.

{'C1': '98.47%', 'C2': '1.53%'}

{'F10': 'ssc_p', 'F3': 'hsc_p', 'F5': 'degree_p', 'F7': 'gender', 'F4': 'degree_t', 'F6': 'workex', 'F1': 'specialisation', 'F8': 'etest_p', 'F2': 'hsc_s', 'F11': 'hsc_b', 'F12': 'ssc_b', 'F9': 'mba_p'}

{'F1': 'F10', 'F2': 'F3', 'F3': 'F5', 'F6': 'F7', 'F10': 'F4', 'F11': 'F6', 'F12': 'F1', 'F4': 'F8', 'F9': 'F2', 'F8': 'F11', 'F7': 'F12', 'F5': 'F9'}

{'C1': 'C1', 'C2': 'C2'}

Not Placed

{'C1': 'Not Placed', 'C2': 'Placed'}

GradientBoostingClassifier

C1

Health Care Services Satisfaction Prediction

Based on the information provided to the classifier, the true label for the given case is likely C1, with a confidence level of 76.26%. Each input variable has a different degree of influence on the classifier's final labelling decision with respect to the case under consideration. Whilst F10, F8, and F7 have lower contributions to the classifier's decision, F16, F15, and F5 are identified as the major contributors resulting in the assignment and classification probabilities across the two classes. There is a 23.74% chance that perhaps C2 is the true label and the features responsible for this are the negative features, F5, F11, F1, F2, F14, F10, and F8. Driving the classifier's decision in favour of C1 are the positive features such as F16, F15, F3, F12, F9, F4, and F6.

{'C2': '23.74%', 'C1': '76.26%'}

{'F16': 'Exact diagnosis', 'F15': 'avaliablity of drugs', 'F5': 'lab services', 'F3': 'friendly health care workers', 'F12': 'Communication with dr', 'F11': 'Time waiting', 'F9': 'Specialists avaliable', 'F4': 'Modern equipment', 'F6': 'waiting rooms', 'F1': 'Check up appointment', 'F13': 'Hygiene and cleaning', 'F2': 'Admin procedures', 'F14': 'Time of appointment', 'F10': 'hospital rooms quality', 'F8': 'parking, playing rooms, caffes', 'F7': 'Quality\\/experience dr.'}

{'F9': 'F16', 'F13': 'F15', 'F12': 'F5', 'F11': 'F3', 'F8': 'F12', 'F2': 'F11', 'F7': 'F9', 'F10': 'F4', 'F14': 'F6', 'F1': 'F1', 'F4': 'F13', 'F3': 'F2', 'F5': 'F14', 'F15': 'F10', 'F16': 'F8', 'F6': 'F7'}

{'C1': 'C2', 'C2': 'C1'}

Satisfied

{'C2': 'Dissatisfied', 'C1': 'Satisfied'}

SGDClassifier

C2

Flight Price-Range Classification

According to the model, C3 is the least probable class, while the most probable class for the given case is identified as C2. The top two variables with the greatest control over the model in terms of this case's label assignment are F11 and F4 but on the contrary, the rest of the variables have moderate-to-lower influence. The contribution of F4 is negative, reducing the chances of selecting the label C2. F11, F2, and F1 drive the model to classify the given case as C2. Furthermore, both F7 and F6 have values that increase the predicted probability of C2, but F9 and F3 decrease the model's response in favour of any of the remaining classes. When choosing a label in this instance, the model pays little attention to the respective values of F8, F12, F5, and F10 hence they are the least ranked features.

{'C2': '86.54%', 'C1': '13.46%', 'C3': '0.00%'}

{'F11': 'Airline', 'F4': 'Total_Stops', 'F2': 'Source', 'F1': 'Journey_month', 'F9': 'Arrival_minute', 'F7': 'Journey_day', 'F6': 'Duration_hours', 'F3': 'Dep_hour', 'F8': 'Destination', 'F12': 'Arrival_hour', 'F5': 'Dep_minute', 'F10': 'Duration_mins'}

{'F9': 'F11', 'F12': 'F4', 'F10': 'F2', 'F2': 'F1', 'F6': 'F9', 'F1': 'F7', 'F7': 'F6', 'F3': 'F3', 'F11': 'F8', 'F5': 'F12', 'F4': 'F5', 'F8': 'F10'}

{'C1': 'C2', 'C2': 'C1', 'C3': 'C3'}

Low

{'C2': 'Low', 'C1': 'Moderate', 'C3': 'High'}

SVC

C1

Australian Credit Approval

Judging by the prediction probabilities, the most probable or likely class assigned by the classifier is C1, with the associated confidence level of 90.97%. The features with the most influence on the prediction above include F9, F13, and F14, while the least important features are F10, F3, and F6. Beside some of the features are shown to negatively contribute to the prediction made here and these negative features, F5, F13, F2, F4, and F10, reduce the classifier's response to generating label C1, consequently pushing the verdict towards C2. The joint impact of the negatives is smaller compared to that of positive features such as F9, F14, F8, and F1, hence the greater drive on the classifier to assign C1 as the correct label.

{'C2': '9.03%', 'C1': '90.97%'}

{'F9': 'A8', 'F13': 'A9', 'F14': 'A12', 'F5': 'A10', 'F8': 'A4', 'F2': 'A14', 'F4': 'A11', 'F1': 'A13', 'F11': 'A1', 'F7': 'A6', 'F12': 'A3', 'F10': 'A5', 'F3': 'A2', 'F6': 'A7'}

{'F8': 'F9', 'F9': 'F13', 'F12': 'F14', 'F10': 'F5', 'F4': 'F8', 'F14': 'F2', 'F11': 'F4', 'F13': 'F1', 'F1': 'F11', 'F6': 'F7', 'F3': 'F12', 'F5': 'F10', 'F2': 'F3', 'F7': 'F6'}

{'C1': 'C2', 'C2': 'C1'}

Class 2

{'C2': 'Class 1', 'C1': 'Class 2'}

KNeighborsClassifier

C1

Credit Risk Classification

According to the model employed, the label for the case is more likely to be C1. This assessment decision is mainly based on the inpacts of features such as F2, F7, F4, F3, and F8. Among these top features, F2, F7, and F4 have positive contributions to the prediction above, while F8 and F3 are identified as negative features which decreases the likelihood associated with class C1 for this case. Furthermore, the values of F10, F9, F5, and F6 also indicate that the other label, C2, may be the correct label but luckily, the influence of the above-mentioned negative features can be classified as only moderate when compared to F2, F7, and F4. In conclusion, with such a strong positive influence from F2, F7, F4, and F1, it is safe to say that the model is very accurate in its classification judgments, with 100.0% certainty.

{'C1': '100.00%', 'C2': '0.00%'}

{'F2': 'fea_4', 'F7': 'fea_8', 'F4': 'fea_2', 'F8': 'fea_9', 'F3': 'fea_6', 'F5': 'fea_10', 'F6': 'fea_1', 'F1': 'fea_7', 'F9': 'fea_11', 'F10': 'fea_3', 'F11': 'fea_5'}

{'F4': 'F2', 'F8': 'F7', 'F2': 'F4', 'F9': 'F8', 'F6': 'F3', 'F10': 'F5', 'F1': 'F6', 'F7': 'F1', 'F11': 'F9', 'F3': 'F10', 'F5': 'F11'}

{'C1': 'C1', 'C2': 'C2'}

Low

{'C1': 'Low', 'C2': 'High'}

BernoulliNB

C2

Personal Loan Modelling

From the prediction likelihood of each class label, the most probable label for the given case based on the values of its features is C2. The likelihood of C1 is negligible, hence we can conclude that the classifier is very confident that C2 is the correct label. Analysing the attributions of the input features showed that the most relevant feature with a strong influence on the classifier's decision here is F4. However, the classifier likely disregards the values of the irrelevant features, F6 and F9, when arriving at the classification above. The confidence level of the classifier employed to make the classification decision above is higher, mainly because the majority of the influential features have positive contributions. Positive features such as F4, F1, and F8 increase the classifier's response higher in favour of C2. F3 and F5 are the main negative features, but compared to F4, their influence on the above classification is very small.

{'C2': '99.99%', 'C1': '0.01%'}

{'F4': 'CD Account', 'F3': 'Income', 'F8': 'CCAvg', 'F1': 'Securities Account', 'F5': 'Education', 'F2': 'Family', 'F7': 'Mortgage', 'F6': 'Age', 'F9': 'Extra_service'}

{'F8': 'F4', 'F2': 'F3', 'F4': 'F8', 'F7': 'F1', 'F5': 'F5', 'F3': 'F2', 'F6': 'F7', 'F1': 'F6', 'F9': 'F9'}

{'C1': 'C2', 'C2': 'C1'}

Reject

{'C2': 'Reject', 'C1': 'Accept'}

LogisticRegression

C2

Bike Sharing Demand

The correct label for the given data instance, according to the machine learning algorithm, is C2 and this is mainly because the probability that C1 is the right label is only about 3.08%. From the analysis, the ranking of the input features based on their respective degree of influence is F3, F11, F12, F5, F2, F10, F8, F9, F6, F7, F4, and F1. This implies the most relevant features are F3, and F11 whereas F4 and F1 are the least relevant ones. Given that F6, F7, and F4 are the features that have a negative impact on the algorithm's selection in this case, it's no wonder that it's quite confident in the chosen class. The arguement towards labelling the case as C2 is also supported by the fact that the joint negative contributions of F6, F7, and F4 is very small when compared to that of the top positive features F3, F11, F12, F5, and F2.

{'C1': '3.08%', 'C2': '96.92%'}

{'F3': 'Functioning Day', 'F11': 'Rainfall(mm)', 'F12': 'Snowfall (cm)', 'F5': 'Solar Radiation (MJ\\/m2)', 'F2': 'Temperature', 'F10': 'Holiday', 'F8': 'Humidity(%)', 'F9': 'Seasons', 'F6': 'Hour', 'F7': 'Visibility (10m)', 'F4': 'Dew point temperature', 'F1': 'Wind speed (m\\/s)'}

{'F12': 'F3', 'F8': 'F11', 'F9': 'F12', 'F7': 'F5', 'F2': 'F2', 'F11': 'F10', 'F3': 'F8', 'F10': 'F9', 'F1': 'F6', 'F5': 'F7', 'F6': 'F4', 'F4': 'F1'}

{'C1': 'C1', 'C2': 'C2'}

More than 500

{'C1': 'Less than 500', 'C2': 'More than 500'}

MLPClassifier

C2

Hotel Satisfaction

Based on the values of the input variables, the prediction model labels the case given as C2 with very high certainty. Specifically, there is only about a 5.59% possibility that C1 is the correct label according to the model. The most influential factors leading to the above prediction decision are the values of F10, F2, and F14 whereas F9, F3, and F6 are deemed less relevant by the model. In between the two ends (most influential and least influential) are the features such as F15, F8, and F11 with moderate contributions. According to the attribution investigation performed, F10, F8, F11, F4, F5, and F6 have positive contributions, increasing the model's response to favour labelling the case as "C2". Conversely, features such as F2, F14, F12, and F15 provide negative contributions, resulting in a small shift toward selecting C1 as the correct class. In conclusion, given that the prediction likelihood of C1 is only 5.59%, it is obvious that the positive features outweigh the negative ones in terms of the considerations they receive from the model, hence the model's decision to assign the C2 label.

{'C2': '94.41%', 'C1': '5.59%'}

{'F10': 'Hotel wifi service', 'F2': 'Type of Travel', 'F14': 'Other service', 'F15': 'Stay comfort', 'F8': 'Type Of Booking', 'F11': 'Ease of Online booking', 'F12': 'Checkin\\/Checkout service', 'F4': 'Age', 'F5': 'Cleanliness', 'F1': 'Food and drink', 'F13': 'Hotel location', 'F7': 'Departure\\/Arrival convenience', 'F9': 'Gender', 'F3': 'purpose_of_travel', 'F6': 'Common Room entertainment'}

{'F6': 'F10', 'F3': 'F2', 'F14': 'F14', 'F11': 'F15', 'F4': 'F8', 'F8': 'F11', 'F13': 'F12', 'F5': 'F4', 'F15': 'F5', 'F10': 'F1', 'F9': 'F13', 'F7': 'F7', 'F1': 'F9', 'F2': 'F3', 'F12': 'F6'}

{'C1': 'C2', 'C2': 'C1'}

dissatisfied

{'C2': 'dissatisfied', 'C1': 'satisfied'}

LogisticRegression

C1

Airline Passenger Satisfaction

The data under consideration is labelled as C1 since it is the most probable label, with a prediction likelihood equal to 99.97% therefore classifier employed here is very confident that C2 is not the right label. The top features with the greatest influence on the classifier in terms of the above classification are F13, F2, F19, and F20. Conversely, the values of F1 and F12 have inconsiderable or insignificant influence on the decision made by the classifier. The input features with moderate to low influence but higher than F1 and F12 on the classifier include F10, F14, F3, and F15. The analysis also shows that the majority of the input features have positive attributions, explaining the level of confidence of the classifier as demonstrated by the prediction probabilities across the classes. The positive features increasing the odds of being labelled C1 include F13, F2, F19, F10, F14, F3, and F17. The marginal doubt in the prediction made here could be attributed to the influence of negative features such as F20, F15, F11, and F9. The negative features support classifying the given data as C2, but since their collective influence is smaller compared to that of the positives, the classifier is shifted more towards labelling the data as C1.

{'C2': '0.03%', 'C1': '99.97%'}

{'F2': 'Inflight wifi service', 'F13': 'Type of Travel', 'F19': 'Customer Type', 'F20': 'Online boarding', 'F10': 'Inflight service', 'F14': 'Baggage handling', 'F3': 'On-board service', 'F15': 'Departure\\/Arrival time convenient', 'F17': 'Seat comfort', 'F11': 'Inflight entertainment', 'F9': 'Gate location', 'F18': 'Cleanliness', 'F22': 'Ease of Online booking', 'F16': 'Class', 'F5': 'Leg room service', 'F4': 'Age', 'F7': 'Departure Delay in Minutes', 'F21': 'Arrival Delay in Minutes', 'F8': 'Gender', 'F6': 'Checkin service', 'F1': 'Food and drink', 'F12': 'Flight Distance'}

{'F7': 'F2', 'F4': 'F13', 'F2': 'F19', 'F12': 'F20', 'F19': 'F10', 'F17': 'F14', 'F15': 'F3', 'F8': 'F15', 'F13': 'F17', 'F14': 'F11', 'F10': 'F9', 'F20': 'F18', 'F9': 'F22', 'F5': 'F16', 'F16': 'F5', 'F3': 'F4', 'F21': 'F7', 'F22': 'F21', 'F1': 'F8', 'F18': 'F6', 'F11': 'F1', 'F6': 'F12'}

{'C1': 'C2', 'C2': 'C1'}

satisfied

{'C2': 'neutral or dissatisfied', 'C1': 'satisfied'}

BernoulliNB

C1

Used Cars Price-Range Prediction

C1 was the predicted category for the given case and the classifier is shown to be very certain about the above prediction verdict, given that the probability of C1 being the label is about 99.72%. The following five features all contributed positively towards the prediction of the C1 class with increasing levels of impact: F4, F5, F10, F6, and F9. F7 and F2 both had similar levels of impact on the prediction of C1, with F7 having a marginally stronger impact. F7 contributed towards the prediction of C1, while F2 contributed against it, in favour of an alternative label. F3 and F1 are the least relevant features, with very little impact, both with negative attributions, driving the prediction decision or verdict away from C1. From the analysis, only the features, F3, F1, and F2, are shown to have negative attributions, shifting the prediction away from C1. However, the collective attribution of F3, F1, and F2 is very low when compared to that of the positive features, so the classifier is motivated strongly by the positive features, leading to the prediction decision above for the case under consideration.

{'C1': '99.72%', 'C2': '0.28%'}

{'F9': 'Transmission', 'F6': 'Fuel_Type', 'F10': 'Seats', 'F5': 'Name', 'F4': 'Engine', 'F7': 'car_age', 'F2': 'Owner_Type', 'F8': 'Power', 'F1': 'Mileage', 'F3': 'Kilometers_Driven'}

{'F8': 'F9', 'F7': 'F6', 'F10': 'F10', 'F6': 'F5', 'F3': 'F4', 'F5': 'F7', 'F9': 'F2', 'F4': 'F8', 'F2': 'F1', 'F1': 'F3'}

{'C1': 'C1', 'C2': 'C2'}

Low

{'C1': 'Low', 'C2': 'High'}

KNeighborsClassifier

C2

Advertisement Prediction

The ML model or algorithm employed here predicted the class C2 with 100.0% confidence level, clearly implying that the case belongs under the class C2 and not C1 since its associated likelihood is 0.0%. Analysis of the contributions of the features indicated that only features F4 and F7 have negative influence, shifting the classification decision away from C2. However, these features are shown to be the least significant ones when it comes to assigning a label to the case under consideration. Therefore, it is a little surprising to see that the model's confidence level is very high with respect to the prediction made here. Among the remaining positive features, F3, and F2, have the strongest impact or influence, increasing the odds of C2 being the label for the case under consideration and the least positive features are F1, F5, and F6.

{'C2': '100.00%', 'C1': '0.00%'}

{'F3': 'Daily Internet Usage', 'F2': 'Daily Time Spent on Site', 'F6': 'Age', 'F1': 'Area Income', 'F5': 'ad_day', 'F7': 'ad_month', 'F4': 'Gender'}

{'F4': 'F3', 'F1': 'F2', 'F2': 'F6', 'F3': 'F1', 'F7': 'F5', 'F6': 'F7', 'F5': 'F4'}

{'C1': 'C2', 'C2': 'C1'}

Skip

{'C2': 'Skip', 'C1': 'Watch'}

BernoulliNB

C2

Hotel Satisfaction

Judging from the values of the input variables, the label predicted for the case under consideration is C2 with a high confidence level of 98.89%, implying that the probability of C1 being the actual label is just 1.11%. The attribution analysis suggests that F11, F10, and F3 are the most impactful features controlling the label selection. In contrast, F8, F13, and F1 are the least important variables whose values contribute marginally to the label selection. While the variables F11, F13, F6, and F9 contribute towards labelling the given case as C1, the remaining variables such as F10, F3, and F15 strongly support the C2 selection. The variables supporting the assignment of C2 are the positive variables whereas negative variables are those shifting the decision in favour of C1 and are against the C2 labelling decision.

{'C2': '98.89%', 'C1': '1.11%'}

{'F11': 'Type of Travel', 'F10': 'Type Of Booking', 'F3': 'Common Room entertainment', 'F15': 'Stay comfort', 'F5': 'Cleanliness', 'F4': 'Hotel wifi service', 'F9': 'Other service', 'F2': 'Ease of Online booking', 'F7': 'Age', 'F6': 'Checkin\\/Checkout service', 'F14': 'Food and drink', 'F12': 'Departure\\/Arrival convenience', 'F8': 'purpose_of_travel', 'F13': 'Hotel location', 'F1': 'Gender'}

{'F3': 'F11', 'F4': 'F10', 'F12': 'F3', 'F11': 'F15', 'F15': 'F5', 'F6': 'F4', 'F14': 'F9', 'F8': 'F2', 'F5': 'F7', 'F13': 'F6', 'F10': 'F14', 'F7': 'F12', 'F2': 'F8', 'F9': 'F13', 'F1': 'F1'}

{'C1': 'C2', 'C2': 'C1'}

dissatisfied

{'C2': 'dissatisfied', 'C1': 'satisfied'}

RandomForestClassifier

C3

Flight Price-Range Classification

There is little to no doubt that C3, among the three classes, is the proper label for this example since its associated predicted probability is 100.0%. F5, F12, and F4 are the variables with the most influence on the labelling output produced here. Furthermore, these variables have a stronger positive influence on the C3 prediction. Similarly, F8, F9, F6, F7, and F10 are some of the variables favouring the selection of C3 as the correct label. F2, F3, and F1, on the other hand, have a negative and opposing impact on the model, increasing the odds in favour of the other labels. When compared to F4, F5, and F12, all of these negative variables have a moderately low impact on the prediction given here. Finally, the lowest ranked essential input variable is recognised as F11, with a very low positive attribution.

{'C3': '100.00%', 'C2': '0.00%', 'C1': '0.00%'}

{'F12': 'Duration_hours', 'F4': 'Airline', 'F5': 'Total_Stops', 'F8': 'Journey_day', 'F2': 'Source', 'F9': 'Destination', 'F6': 'Journey_month', 'F7': 'Dep_minute', 'F10': 'Arrival_minute', 'F3': 'Arrival_hour', 'F1': 'Duration_mins', 'F11': 'Dep_hour'}

{'F7': 'F12', 'F9': 'F4', 'F12': 'F5', 'F1': 'F8', 'F10': 'F2', 'F11': 'F9', 'F2': 'F6', 'F4': 'F7', 'F6': 'F10', 'F5': 'F3', 'F8': 'F1', 'F3': 'F11'}

{'C1': 'C3', 'C2': 'C2', 'C3': 'C1'}

Low

{'C3': 'Low', 'C2': 'Moderate', 'C1': 'High'}

BernoulliNB

C1

Student Job Placement

Here, the model assigned C1 the highest probability, equal to 99.48%, implying that the predictability of C2 is only 0.52%. Per the attribution analysis, only F1 and F9 have negative contributions that decrease the likelihood of the C1 label in favour of the C2 label. F12, F10, F8, and F2 have the highest positive contributions that improve the odds in favour of the C1. The contributions of the other positive features, such as F4, F6, and F7, have moderate contributions, whilst F3, F11, and F5 are the lowest ranked positive features. All in all, the model is very certain that C2 is not the true label, and this highlighted by the fact that the joint negative contribution of F9 and F1 is only marginal when compared with the very strong influence of positive features such as F12, F8, and F2.

{'C2': '0.52%', 'C1': '99.48%'}

{'F12': 'workex', 'F8': 'specialisation', 'F2': 'ssc_p', 'F10': 'hsc_p', 'F1': 'degree_p', 'F4': 'gender', 'F6': 'degree_t', 'F7': 'etest_p', 'F3': 'hsc_b', 'F9': 'hsc_s', 'F11': 'ssc_b', 'F5': 'mba_p'}

{'F11': 'F12', 'F12': 'F8', 'F1': 'F2', 'F2': 'F10', 'F3': 'F1', 'F6': 'F4', 'F10': 'F6', 'F4': 'F7', 'F8': 'F3', 'F9': 'F9', 'F7': 'F11', 'F5': 'F5'}

{'C1': 'C2', 'C2': 'C1'}

Placed

{'C2': 'Not Placed', 'C1': 'Placed'}

RandomForestClassifier

C2

Employee Attrition

The assigned label or class by the prediction algorithm is C2, which happens to be the most probable class predicted with a probability of around 56.0%, consequently, there is a 44.0% chance that perhaps C1 could be the true label instead. The classification assertion above is attributed to the contributions of mainly F23, F1, F18, F12, F30, F10, F11, F16, F29, F24, F8, F15, F7, F28, F9, F17, F14, F6, F26, and F25. However, not all of the features are considered relevant when determining the correct label for the given case. F5, F4, F27, and F2 are examples of irrelevant features. Among the influential features, F23 and F1 are regarded as the most negative, dragging the verdict in a different direction, while the top features, F18 and F12, have positive contributions, increasing the likelihood that C1 is the right label here. Actually, the reason for the 44.0% prediction likelihood of C1 can be attributed to the strong negative influence of F23 and F1. The other negative features include F30, F10, and F29, while the other positive features are F11, F16, F24, and F8.

{'C1': '44.00%', 'C2': '56.00%'}

{'F23': 'OverTime', 'F1': 'BusinessTravel', 'F18': 'MaritalStatus', 'F12': 'JobInvolvement', 'F30': 'WorkLifeBalance', 'F10': 'Education', 'F11': 'EnvironmentSatisfaction', 'F16': 'Gender', 'F29': 'JobRole', 'F24': 'NumCompaniesWorked', 'F8': 'YearsInCurrentRole', 'F15': 'HourlyRate', 'F7': 'Department', 'F28': 'RelationshipSatisfaction', 'F9': 'PerformanceRating', 'F17': 'YearsWithCurrManager', 'F14': 'Age', 'F6': 'MonthlyRate', 'F25': 'StockOptionLevel', 'F26': 'JobSatisfaction', 'F5': 'DailyRate', 'F2': 'YearsSinceLastPromotion', 'F4': 'YearsAtCompany', 'F27': 'TrainingTimesLastYear', 'F21': 'EducationField', 'F13': 'TotalWorkingYears', 'F20': 'PercentSalaryHike', 'F22': 'MonthlyIncome', 'F3': 'JobLevel', 'F19': 'DistanceFromHome'}

{'F26': 'F23', 'F17': 'F1', 'F25': 'F18', 'F29': 'F12', 'F20': 'F30', 'F27': 'F10', 'F28': 'F11', 'F23': 'F16', 'F24': 'F29', 'F8': 'F24', 'F14': 'F8', 'F4': 'F15', 'F21': 'F7', 'F18': 'F28', 'F19': 'F9', 'F16': 'F17', 'F1': 'F14', 'F7': 'F6', 'F10': 'F25', 'F30': 'F26', 'F2': 'F5', 'F15': 'F2', 'F13': 'F4', 'F12': 'F27', 'F22': 'F21', 'F11': 'F13', 'F9': 'F20', 'F6': 'F22', 'F5': 'F3', 'F3': 'F19'}

{'C1': 'C1', 'C2': 'C2'}

Leave

{'C1': 'Leave', 'C2': 'Leave'}

GradientBoostingClassifier

C1

Broadband Sevice Signup

For the given case, the model predicts C1 as the label. The probability that the label could be the alternative class, C2, is only about 1.94% which implies that the model is very confident in this classification decision or output. F5 and F1 are the top features pushing for the C1 prediction for this case. Other features with a positive impact on this prediction include F22, F6, F12, F29, and F11. On the other hand, the values of F17, F35, F15, and F14 make up the set of features with negative attributions on the prediction decision above. However, compared to F9, F22, F6, and F1, the features above have a very marginal influence on the model. This might explain why the model is highly confident that the true label is likely C1. Finally, there were some features with insignificant impact on the model's prediction decision for the case under consideration and these include F21, F36, F42, and F13.

{'C1': '98.06%', 'C2': '1.94%'}

{'F9': 'X38', 'F1': 'X22', 'F22': 'X32', 'F6': 'X19', 'F12': 'X1', 'F11': 'X13', 'F17': 'X11', 'F29': 'X3', 'F35': 'X16', 'F15': 'X2', 'F14': 'X12', 'F7': 'X14', 'F2': 'X42', 'F18': 'X18', 'F33': 'X28', 'F5': 'X35', 'F37': 'X24', 'F19': 'X20', 'F24': 'X8', 'F40': 'X40', 'F21': 'X34', 'F36': 'X5', 'F13': 'X4', 'F42': 'X41', 'F27': 'X6', 'F39': 'X39', 'F10': 'X7', 'F20': 'X37', 'F16': 'X36', 'F28': 'X33', 'F34': 'X21', 'F26': 'X9', 'F23': 'X31', 'F32': 'X30', 'F38': 'X10', 'F31': 'X27', 'F30': 'X26', 'F4': 'X25', 'F8': 'X15', 'F25': 'X23', 'F41': 'X17', 'F3': 'X29'}

{'F35': 'F9', 'F20': 'F1', 'F29': 'F22', 'F17': 'F6', 'F40': 'F12', 'F11': 'F11', 'F9': 'F17', 'F2': 'F29', 'F14': 'F35', 'F1': 'F15', 'F10': 'F14', 'F12': 'F7', 'F38': 'F2', 'F16': 'F18', 'F26': 'F33', 'F32': 'F5', 'F22': 'F37', 'F18': 'F19', 'F6': 'F24', 'F37': 'F40', 'F31': 'F21', 'F41': 'F36', 'F3': 'F13', 'F39': 'F42', 'F4': 'F27', 'F36': 'F39', 'F5': 'F10', 'F34': 'F20', 'F33': 'F16', 'F30': 'F28', 'F19': 'F34', 'F7': 'F26', 'F28': 'F23', 'F27': 'F32', 'F8': 'F38', 'F25': 'F31', 'F24': 'F30', 'F23': 'F4', 'F13': 'F8', 'F21': 'F25', 'F15': 'F41', 'F42': 'F3'}

{'C1': 'C1', 'C2': 'C2'}

No

{'C1': 'No', 'C2': 'Yes'}

RandomForestClassifier

C1

E-Commerce Shipping

The predicted likelihood of C1 based on the information supplied to the model is 51.62%, whereas there is a 48.38% likelihood that C2 is the correct label. The uncertainty of the model in terms of this case or instance can be attributed mainly to the direction of influence of the variables F5, F2, and F8. Decreasing the chances of C1 being the correct label are the variables F5, F8, F7, and F4. While F5, F8, and F7 have strong negative attributions, F4 is the least negative variable. Increasing the likelihood of C1 prediction are mainly the variables F2, F1, and F9. The features F3, F6, and F10 also have a weak positive influence on the classification decision arrived at by the model for this case under consideration.

{'C1': '51.62%', 'C2': '48.38%'}

{'F5': 'Discount_offered', 'F2': 'Weight_in_gms', 'F8': 'Customer_care_calls', 'F7': 'Product_importance', 'F9': 'Mode_of_Shipment', 'F1': 'Warehouse_block', 'F10': 'Cost_of_the_Product', 'F6': 'Gender', 'F3': 'Customer_rating', 'F4': 'Prior_purchases'}

{'F2': 'F5', 'F3': 'F2', 'F6': 'F8', 'F9': 'F7', 'F5': 'F9', 'F4': 'F1', 'F1': 'F10', 'F10': 'F6', 'F7': 'F3', 'F8': 'F4'}

{'C1': 'C1', 'C2': 'C2'}

On-time

{'C1': 'On-time', 'C2': 'Late'}

SVC

C1

Tic-Tac-Toe Strategy

In this case, the classifier indicates that there is a 99.50% chance that the C1 class is the true label, so it is correct to conclude that the classifier is not sure that C2 is the correct label for the case here. According to the study, five input variables contradict the label choice, while four variables support the classification made above. The variables that contradict the prediction are known as negative features while positive features are those that support the classification verdict. F1, F2, F7, F8, and F5 are the negative variables that reduce the likelihood of C1 being the correct label. F3, F6, F9, and F4 are the positive variables that increase the likelihood of C1.

{'C2': '0.50%', 'C1': '99.50%'}

{'F1': 'middle-middle-square', 'F3': 'top-left-square', 'F6': 'bottom-left-square', 'F9': 'bottom-right-square', 'F2': ' top-right-square', 'F7': 'middle-right-square', 'F8': 'top-middle-square', 'F4': 'middle-left-square', 'F5': 'bottom-middle-square'}

{'F5': 'F1', 'F1': 'F3', 'F7': 'F6', 'F9': 'F9', 'F3': 'F2', 'F6': 'F7', 'F2': 'F8', 'F4': 'F4', 'F8': 'F5'}

{'C1': 'C2', 'C2': 'C1'}

player B win

{'C2': 'player B lose', 'C1': 'player B win'}

LogisticRegression

C1

Employee Promotion Prediction

Considering the values of features such as F9, F8, and F11, the model is very certain (about 99.65% certain) that C1 is the right label for the given case. While F9, F8, and F11 are the most important features, the model paid little attention to F1, F5, and F7 when deciding on the appropriate label here.Overall, driving down the odds of C1 are the negative features F11, F6, F3, and F5, which are shown to support the other label. However, the very high confidence in the above-mentioned decision is chiefly attributed to the positive contributions of F8, F9, F10, F4, and F2.

{'C1': '99.65%', 'C2': '0.35%'}

{'F8': 'avg_training_score', 'F9': 'KPIs_met >80%', 'F11': 'department', 'F10': 'age', 'F4': 'gender', 'F2': 'region', 'F6': 'length_of_service', 'F3': 'recruitment_channel', 'F1': 'previous_year_rating', 'F5': 'no_of_trainings', 'F7': 'education'}

{'F11': 'F8', 'F10': 'F9', 'F1': 'F11', 'F7': 'F10', 'F4': 'F4', 'F2': 'F2', 'F9': 'F6', 'F5': 'F3', 'F8': 'F1', 'F6': 'F5', 'F3': 'F7'}

{'C1': 'C1', 'C2': 'C2'}

Ignore

{'C1': 'Ignore', 'C2': 'Promote'}

KNeighborsClassifier

C2

Suspicious Bidding Identification

With a certainty of 100.0%, the model labels this case as C2 and from the predicted likelihoods across the classes, it can be inferred that the model verdict is that there is a zero chance that the case is under C1. The most significant feature is F1, while the least important attributes are F8, F7, and F3. The moderate features are F4, F9, F6, F5, and F2, ranked in order of their respective attributions on the label predicted. With regards to the direction of influence of each feature, some of the input features have positive attributions in favour of the assigned label and increasing the response of the model in favour of the C2 label, while the remaining ones contradict. F1, F6, F5, and F8 are the positive features, while F4, F9, F2, F7, and F3 are the negative ones, shifting the prediction verdict in the direction of C1.

{'C2': '100.00%', 'C1': '0.00%'}

{'F1': 'Z3', 'F4': 'Z9', 'F9': 'Z4', 'F6': 'Z8', 'F5': 'Z1', 'F2': 'Z5', 'F8': 'Z2', 'F7': 'Z6', 'F3': 'Z7'}

{'F3': 'F1', 'F9': 'F4', 'F4': 'F9', 'F8': 'F6', 'F1': 'F5', 'F5': 'F2', 'F2': 'F8', 'F6': 'F7', 'F7': 'F3'}

{'C1': 'C2', 'C2': 'C1'}

Normal

{'C2': 'Normal', 'C1': 'Suspicious'}

BernoulliNB

C2

Cab Surge Pricing System

The case under consideration can be labelled as either C2 or C1 or C3, and based on values for features such as F1, F6, F12, F4, and F5, the model labelled this test case as C2 with a confidence level equal to 62.29%. However, there is a 28.41% chance that the label could be C1 and a 9.3% chance that it could be C3. All the features used to make the prediction decision have different influences on the model with respect to this test case. That is, while some features positively support the prediction, others have values suggesting any of the alternative labels could be the true label. According to the analysis, F1, F12, F6, and F4 are the top features with the highest impact on the prediction made. The features F1, F12, F4, and F6 are the top attributes positively supporting the prediction of C2. In contrast, F5 and F11 are the features with the most negative attributions, pushing for the prediction of an alternative class. Further decreasing the likelihood of C2 are the features F10, F9, F3, and F8, which all negatively contribute to the model's final decision with respect to the given case. Finally, features F2 and F7 are shown to be less relevant, with positive contributions to the above classification.

{'C2': '62.29%', 'C1': '28.41%', 'C3': '9.30%'}

{'F1': 'Confidence_Life_Style_Index', 'F6': 'Destination_Type', 'F12': 'Customer_Rating', 'F4': 'Type_of_Cab', 'F5': 'Cancellation_Last_1Month', 'F11': 'Trip_Distance', 'F10': 'Var1', 'F9': 'Customer_Since_Months', 'F3': 'Gender', 'F8': 'Var3', 'F2': 'Life_Style_Index', 'F7': 'Var2'}

{'F5': 'F1', 'F6': 'F6', 'F7': 'F12', 'F2': 'F4', 'F8': 'F5', 'F1': 'F11', 'F9': 'F10', 'F3': 'F9', 'F12': 'F3', 'F11': 'F8', 'F4': 'F2', 'F10': 'F7'}

{'C1': 'C2', 'C2': 'C1', 'C3': 'C3'}

C1

{'C2': 'Low', 'C1': 'Medium', 'C3': 'High'}

SVC

C1

Advertisement Prediction

Tasked with labelling a given case as either class C1 or class C2 , the model assigns C1 as the most probable true label, with a confidence level of approximately 99.90%. This confidence level suggests that the probability of C2 being the correct label is only 0.10%. Attribution analysis conducted indicates that all the variables have a different degree of influence or contribution to the model arriving at the above mentioned classification verdict. The features responsible for the very high certainty of the model with respect to the case under consideration are F2, F6, F3, and F1. Actually, the only input variables with a negative contribution also happen to be the least relevant variables, F4 and F7.

{'C1': '99.90%', 'C2': '0.10%'}

{'F2': 'Daily Internet Usage', 'F6': 'Daily Time Spent on Site', 'F3': 'Age', 'F1': 'ad_day', 'F5': 'Area Income', 'F4': 'Gender', 'F7': 'ad_month'}

{'F4': 'F2', 'F1': 'F6', 'F2': 'F3', 'F7': 'F1', 'F3': 'F5', 'F5': 'F4', 'F6': 'F7'}

{'C1': 'C1', 'C2': 'C2'}

Skip

{'C1': 'Skip', 'C2': 'Watch'}

SVC

C2

Water Quality Classification

Even though there is moderately high confidence in the assigned label, the prediction probabilities across the two classes indicate that C1 could be the correct label for this data instance. The variables with primary contributions resulting in the labelling decision above are F3, F2, F5, and F4. As per the attribution analysis, the top two variables, F3 and F2, have a negative impact, influencing the classifier to label the given data as C1 instead of C2. The only other negative variable is F6, with moderate influence compared to the other two negative variables. On the other hand, there are many variables, specifically F5, F4, F8, F1, F7, and F9, that positively support and influence the classifier to assign C2. To a greater degree, the level of uncertainty with respect to this classification instance could be explained away by just looking at the negative variables' fairly strong pull on the classifier towards C1.

{'C1': '38.68%', 'C2': '61.32%'}

{'F3': 'Sulfate', 'F2': 'Hardness', 'F5': 'ph', 'F4': 'Conductivity', 'F6': 'Turbidity', 'F8': 'Chloramines', 'F1': 'Solids', 'F7': 'Trihalomethanes', 'F9': 'Organic_carbon'}

{'F5': 'F3', 'F2': 'F2', 'F1': 'F5', 'F6': 'F4', 'F9': 'F6', 'F4': 'F8', 'F3': 'F1', 'F8': 'F7', 'F7': 'F9'}

{'C1': 'C1', 'C2': 'C2'}

Portable

{'C1': 'Not Portable', 'C2': 'Portable'}

RandomForestClassifier

C2

Student Job Placement

According to the classification model employed here, there is a marginal chance that the true label for this test example is C1. Undoubtedly, the model estimated that the likelihood of the true label being equal to C2 is 99.92%. The above prediction decision is based on the influence of features such as F3, F1, F2, F5, and F7. All these features have significant positive support for the prediction decision here, with the top features being F1 and F2. Furthermore, the features with a moderate influence on the prediction of C2 are F6, F8, F9, and F4. While F4 positively supports labelling the case under consideration as C2, the features F6, F8, and F9 indicate otherwise. Finally, the features with marginal impact are F10, F12, and F11.

{'C1': '0.08%', 'C2': '99.92%'}

{'F1': 'ssc_p', 'F2': 'workex', 'F3': 'hsc_p', 'F5': 'specialisation', 'F7': 'gender', 'F6': 'mba_p', 'F8': 'hsc_s', 'F4': 'ssc_b', 'F9': 'degree_t', 'F10': 'hsc_b', 'F11': 'degree_p', 'F12': 'etest_p'}

{'F1': 'F1', 'F11': 'F2', 'F2': 'F3', 'F12': 'F5', 'F6': 'F7', 'F5': 'F6', 'F9': 'F8', 'F7': 'F4', 'F10': 'F9', 'F8': 'F10', 'F3': 'F11', 'F4': 'F12'}

{'C1': 'C1', 'C2': 'C2'}

Placed

{'C1': 'Not Placed', 'C2': 'Placed'}

SVM

C2

Customer Churn Modelling

Considering the values of the features, the prediction from the model for the case under consideration is C2 and this labelling decision is not 100% certain given that there is a 27.27% probability that it could be C1. For the case under consideration, the assigned label is mainly due to the values of the features F3, F9, F7, and F2 while the least important is F1. The direction of the contributions of the relevant features is summarised in the following sentences: F3 and F9 have a very strong joint positive contribution in favour of class C2 coupled with moderately positive input features F7, F2, and F4, however unlike them, F1 has a very low positive impact on the model for the case here. All of F6, F10, F8, and F5 have a negative impact on the prediction made here, however, their pull is not enough to shift the prediction in the direction of the other class label, C1.

{'C1': '27.27%', 'C2': '72.73%'}

{'F3': 'Age', 'F9': 'IsActiveMember', 'F7': 'Geography', 'F2': 'NumOfProducts', 'F4': 'Gender', 'F6': 'Tenure', 'F10': 'CreditScore', 'F8': 'EstimatedSalary', 'F5': 'Balance', 'F1': 'HasCrCard'}

{'F4': 'F3', 'F9': 'F9', 'F2': 'F7', 'F7': 'F2', 'F3': 'F4', 'F5': 'F6', 'F1': 'F10', 'F10': 'F8', 'F6': 'F5', 'F8': 'F1'}

{'C1': 'C1', 'C2': 'C2'}

Leave

{'C1': 'Stay', 'C2': 'Leave'}

DNN

C1

Concrete Strength Classification

For this case, the classification model's confidence is only about 69.40%, implying that the likelihood of label C2 is about 30.60%. According to the classification attribution analysis, F1 and F2 are the most relevant features, whereas F4 and F6 are the least influential. When the attributions of the features were carefully analysed, only F8, F5, and F7 are identified as negative features since their contributions drive down the prediction likelihood of the assigned label, C1. Conversely, F1, F2, F3, F4, and F6 have a positive influence on the model in support of labelling the given case as C1 instead of C2.

{'C1': '69.40%', 'C2': '30.60%'}

{'F1': 'slag', 'F2': 'water', 'F8': 'cement', 'F5': 'fineaggregate', 'F3': 'flyash', 'F7': 'coarseaggregate', 'F4': 'age_days', 'F6': 'superplasticizer'}

{'F2': 'F1', 'F4': 'F2', 'F1': 'F8', 'F7': 'F5', 'F3': 'F3', 'F6': 'F7', 'F8': 'F4', 'F5': 'F6'}

{'C1': 'C1', 'C2': 'C2'}

Weak

{'C1': 'Weak', 'C2': 'Strong'}

LogisticRegression

C2

Hotel Satisfaction

The model prediction for the test case is C2 and the confidence level of this prediction decision is 91.36%, while the predicted probability of C1 is only 8.64%. According to the attribution analysis, we can see that the features F10 and F8 have negative attributions, pushing the prediction decision towards the alternative label, C1. Conversely, the F9, F13, F1, and F2 have values with a positive impact, shifting the classification decision towards label C2. Furthermore, while the attributes F14 and F15 contradict the prediction made, F6 and F12 have values that support the prediction from the model for the test case under consideration. Finally, F7, F3, F5, and F4 are the least ranked features, and among them, only F4 has a negative influence that contributes marginally to the shift away from labelling the case as C2.

{'C2': '91.36%', 'C1': '8.64%'}

{'F10': 'Type of Travel', 'F8': 'Type Of Booking', 'F9': 'Hotel wifi service', 'F13': 'Common Room entertainment', 'F1': 'Stay comfort', 'F2': 'Other service', 'F14': 'Checkin\\/Checkout service', 'F6': 'Hotel location', 'F15': 'Food and drink', 'F12': 'Cleanliness', 'F11': 'Age', 'F7': 'Departure\\/Arrival convenience', 'F3': 'purpose_of_travel', 'F4': 'Ease of Online booking', 'F5': 'Gender'}

{'F3': 'F10', 'F4': 'F8', 'F6': 'F9', 'F12': 'F13', 'F11': 'F1', 'F14': 'F2', 'F13': 'F14', 'F9': 'F6', 'F10': 'F15', 'F15': 'F12', 'F5': 'F11', 'F7': 'F7', 'F2': 'F3', 'F8': 'F4', 'F1': 'F5'}

{'C1': 'C2', 'C2': 'C1'}

dissatisfied

{'C2': 'dissatisfied', 'C1': 'satisfied'}

RandomForestClassifier

C2

Student Job Placement

In summary, the model predicted an 87.14% likelihood of the class label C2 for the test example under consideration, therefore, there is a chance of about 12.86% that the correct class label could be a different label. The features with the highest impact on the model are F7, F3, F4, and F10, whose values are attributing most to the labeling decision here and among these features, only F10 shows the potential to shift the narrative toward a different label. On impact comparison, features F7, F3, F4 and F10 have higher impact on the model prediction than F5 and F9. Features F7, F3, F4, F5, and F9 show a positive impact shifting towards the prediction of C2. F10 is the most negative of all the set of features passed to the model, F1, F8, and F6 have moderate negative influence, whereas the feature F11 has very little negative impact on the prediction.

{'C2': '87.14%', 'C1': '12.86%'}

{'F7': 'ssc_p', 'F3': 'hsc_p', 'F4': 'degree_p', 'F10': 'workex', 'F5': 'specialisation', 'F9': 'gender', 'F12': 'hsc_s', 'F1': 'etest_p', 'F6': 'degree_t', 'F2': 'mba_p', 'F8': 'ssc_b', 'F11': 'hsc_b'}

{'F1': 'F7', 'F2': 'F3', 'F3': 'F4', 'F11': 'F10', 'F12': 'F5', 'F6': 'F9', 'F9': 'F12', 'F4': 'F1', 'F10': 'F6', 'F5': 'F2', 'F7': 'F8', 'F8': 'F11'}

{'C1': 'C2', 'C2': 'C1'}

Not Placed

{'C2': 'Not Placed', 'C1': 'Placed'}

LogisticRegression

C1

Food Ordering Customer Churn Prediction

Mainly based on the values of the features F23, F13, F24, and F30, the model classifies the given case as C1 with a prediction confidence level of 90.15%. This means that there is only a 9.85% chance that the correct label could be C2. The features that positively contribute to the prediction include F23, F30, F43, and F18, since their influences increase the model's response in favour of assigning the label C1. On the flip side, features dragging the final decision higher towards C2 include F13, F24, F27, and F21, since their values contradict the assigned label here. Finally, the prediction was made with less emphasis on the values of features such as F4, F26, F7, and F25, given that they are shown to have very close to zero influence.

{'C2': '9.85%', 'C1': '90.15%'}

{'F23': 'Unaffordable', 'F13': 'Perference(P2)', 'F24': 'Influence of rating', 'F30': 'Good Food quality', 'F43': 'Delay of delivery person picking up food', 'F18': 'Less Delivery time', 'F27': 'Freshness ', 'F5': 'Politeness', 'F21': 'Ease and convenient', 'F31': 'More restaurant choices', 'F6': 'Missing item', 'F39': 'Order Time', 'F38': 'Gender', 'F19': 'Time saving', 'F20': 'Unavailability', 'F8': 'Late Delivery', 'F2': 'Temperature', 'F16': 'High Quality of package', 'F9': 'Long delivery time', 'F33': 'Poor Hygiene', 'F26': 'Low quantity low time', 'F4': 'Delivery person ability', 'F7': 'Number of calls', 'F25': 'Google Maps Accuracy', 'F3': 'Residence in busy location', 'F22': 'Good Taste ', 'F42': 'Maximum wait time', 'F46': 'Influence of time', 'F15': 'Good Road Condition', 'F40': 'Age', 'F44': 'Order placed by mistake', 'F34': 'Wrong order delivered', 'F10': 'Delay of delivery person getting assigned', 'F45': 'Family size', 'F41': 'Bad past experience', 'F17': 'Health Concern', 'F36': 'Self Cooking', 'F12': 'Good Tracking system', 'F14': 'More Offers and Discount', 'F37': 'Easy Payment option', 'F11': 'Perference(P1)', 'F1': 'Educational Qualifications', 'F29': 'Monthly Income', 'F32': 'Occupation', 'F35': 'Marital Status', 'F28': 'Good Quantity'}

{'F23': 'F23', 'F9': 'F13', 'F38': 'F24', 'F15': 'F30', 'F26': 'F43', 'F39': 'F18', 'F43': 'F27', 'F42': 'F5', 'F10': 'F21', 'F12': 'F31', 'F28': 'F6', 'F31': 'F39', 'F2': 'F38', 'F11': 'F19', 'F22': 'F20', 'F19': 'F8', 'F44': 'F2', 'F40': 'F16', 'F24': 'F9', 'F20': 'F33', 'F36': 'F26', 'F37': 'F4', 'F41': 'F7', 'F34': 'F25', 'F33': 'F3', 'F45': 'F22', 'F32': 'F42', 'F30': 'F46', 'F35': 'F15', 'F1': 'F40', 'F29': 'F44', 'F27': 'F34', 'F25': 'F10', 'F7': 'F45', 'F21': 'F41', 'F18': 'F17', 'F17': 'F36', 'F16': 'F12', 'F14': 'F14', 'F13': 'F37', 'F8': 'F11', 'F6': 'F1', 'F5': 'F29', 'F4': 'F32', 'F3': 'F35', 'F46': 'F28'}

{'C1': 'C2', 'C2': 'C1'}

Go Away

{'C2': 'Return', 'C1': 'Go Away'}

SVC

C2

Bike Sharing Demand

90.58% it the predicted chance that C2 is the correct label for the given case, indicating that the predicted probability of C1 is only 9.42%. Per the feature-attributions, the top-ranked features are F12, F7, and F6, whereas the smallest important or least ranked features are F3, F11, F10, and F9. The influence of intermediate input features like F4, F5, and F2 is considered moderate. The features with positive contributions to the classification above are F6, F5, F3, and F11, while on the other hand, all the remaining features are shown to negatively contribute to the decision above. The main negative features that decrease the probability that C2 is the true label, considering the likelihood of label C1 for this case, are F12, F7, and F4.

{'C2': '90.58%', 'C1': '9.42%'}

{'F12': 'Functioning Day', 'F6': 'Solar Radiation (MJ\\/m2)', 'F7': 'Rainfall(mm)', 'F4': 'Snowfall (cm)', 'F5': 'Hour', 'F2': 'Temperature', 'F1': 'Holiday', 'F8': 'Humidity(%)', 'F3': 'Visibility (10m)', 'F11': 'Dew point temperature', 'F10': 'Seasons', 'F9': 'Wind speed (m\\/s)'}

{'F12': 'F12', 'F7': 'F6', 'F8': 'F7', 'F9': 'F4', 'F1': 'F5', 'F2': 'F2', 'F11': 'F1', 'F3': 'F8', 'F5': 'F3', 'F6': 'F11', 'F10': 'F10', 'F4': 'F9'}

{'C1': 'C2', 'C2': 'C1'}

Less than 500

{'C2': 'Less than 500', 'C1': 'More than 500'}

GradientBoostingClassifier

C1

Paris House Classification

The most likely label for the given scenario, according to this prediction, is C1, which has a prediction probability of 97.02 percent, whereas C2 has a prediction probability of just 2.98 percent. The impact of F14, F7, and F1 is mostly responsible for the aforementioned classification. F13, F12, and F17 are the following groups of features with moderate contributions. F16, F9, F2, and F8, on the other hand, receive minimal attention from the classifier. Given that all four top features have a substantial positive contribution, it's easy to see why the classifier is quite certain that C1 is the correct label in this case. F12, F11, and F6 are also negative features, having a moderate to low influence.

{'C2': '2.98%', 'C1': '97.02%'}

{'F14': 'isNewBuilt', 'F7': 'hasYard', 'F1': 'hasPool', 'F13': 'hasStormProtector', 'F12': 'made', 'F17': 'squareMeters', 'F11': 'floors', 'F6': 'cityCode', 'F4': 'hasGuestRoom', 'F5': 'basement', 'F15': 'numPrevOwners', 'F10': 'price', 'F3': 'numberOfRooms', 'F16': 'garage', 'F9': 'cityPartRange', 'F2': 'hasStorageRoom', 'F8': 'attic'}

{'F3': 'F14', 'F1': 'F7', 'F2': 'F1', 'F4': 'F13', 'F12': 'F12', 'F6': 'F17', 'F8': 'F11', 'F9': 'F6', 'F16': 'F4', 'F13': 'F5', 'F11': 'F15', 'F17': 'F10', 'F7': 'F3', 'F15': 'F16', 'F10': 'F9', 'F5': 'F2', 'F14': 'F8'}

{'C1': 'C2', 'C2': 'C1'}

Luxury

{'C2': 'Basic', 'C1': 'Luxury'}

GradientBoostingClassifier

C2

German Credit Evaluation

According to the prediction algorithm employed here, the most probable label for the given data instance is C2. The confidence level associated with the prediction decision above is 64.62%, meaning there is about a 35.38% likelihood that C1 is the right choice. The input features can be ranked according to their respective degrees of influence in decreasing order as follows: F4, F3, F1, F8, F2, F9, F5, F7, and F6. Therefore, when classifying the given case, the algorithm places little emphasis or consideration on the values of F3 and F4, however, the values of F7 and F6 are the most important here. F6, F9, F5, and F1 are regarded as negative features since their contributions decrease the likelihood of C2 being the correct label. However, positive features such as F7, F8, and F2 drive the algorithm higher towards assigning C2 to the case under consideration here.

{'C2': '64.62%', 'C1': '35.38%'}

{'F6': 'Saving accounts', 'F7': 'Sex', 'F5': 'Duration', 'F9': 'Housing', 'F2': 'Checking account', 'F8': 'Purpose', 'F1': 'Credit amount', 'F3': 'Age', 'F4': 'Job'}

{'F5': 'F6', 'F2': 'F7', 'F8': 'F5', 'F4': 'F9', 'F6': 'F2', 'F9': 'F8', 'F7': 'F1', 'F1': 'F3', 'F3': 'F4'}

{'C1': 'C2', 'C2': 'C1'}

Good Credit

{'C2': 'Good Credit', 'C1': 'Bad Credit'}

KNeighborsClassifier

C2

Tic-Tac-Toe Strategy

The true label has a 50.0% chance of being one of the two classes and based on the predicted likelihoods mentioned above, it can be concluded that the model is very unsure about the correctness of the classification. The above prediction decisions are mainly influenced by the features F2, F3, F4, F1, F5, and F7, while the least important are F8, F6, and F9. Overall, since the predicted likelihood is evenly split between the two classes, it can be concluded that the model is very uncertain as to which label is the right one. The variables with contributions that support the assignment of C2 include F2, F5, F6, and F9, but on the other hand, the ones with contributions towards the assignment of C1 are F3, F7, F4, F1, and F8. With respect to the assignment of the C2 label, F3, F7, F4, F1, and F8 are the negative variables, while F2, F5, F6, and F9 are the positive variables.

{'C2': '50.00%', 'C1': '50.00%'}

{'F2': 'middle-middle-square', 'F3': 'top-left-square', 'F7': 'bottom-left-square', 'F4': 'bottom-right-square', 'F1': 'top-middle-square', 'F5': ' top-right-square', 'F8': 'middle-right-square', 'F6': 'bottom-middle-square', 'F9': 'middle-left-square'}

{'F5': 'F2', 'F1': 'F3', 'F7': 'F7', 'F9': 'F4', 'F2': 'F1', 'F3': 'F5', 'F6': 'F8', 'F8': 'F6', 'F4': 'F9'}

{'C1': 'C2', 'C2': 'C1'}

player B lose

{'C2': 'player B lose', 'C1': 'player B win'}

DecisionTreeClassifier

C1

Credit Risk Classification

The model is assigned the label C1 for the given example. F10, F2, and F4 are the most important features that influence the above-mentioned estimate decision, however unlike them, F6, F1, and F11 are less important. The majority of features have values that swing the judgement towards the other label, C2. The only input features that increase the likelihood that C1 is the correct label are F10, F5, and F1, therefore it is very surprising that the model has 100.0% confidence in its estimate for the given example.

{'C2': '0.00%', 'C1': '100.00%'}

{'F10': 'fea_4', 'F2': 'fea_8', 'F4': 'fea_5', 'F9': 'fea_2', 'F3': 'fea_1', 'F8': 'fea_9', 'F7': 'fea_11', 'F5': 'fea_6', 'F6': 'fea_10', 'F1': 'fea_7', 'F11': 'fea_3'}

{'F4': 'F10', 'F8': 'F2', 'F5': 'F4', 'F2': 'F9', 'F1': 'F3', 'F9': 'F8', 'F11': 'F7', 'F6': 'F5', 'F10': 'F6', 'F7': 'F1', 'F3': 'F11'}

{'C1': 'C2', 'C2': 'C1'}

High

{'C2': 'Low', 'C1': 'High'}

RandomForestClassifier

C1

Annual Income Earnings

Since the probability that C2 is the correct label is only 2.18%, the classifier assigns the label C1 in this labelling instance. The main factors influencing this classification decision are the values of the variables F4, F11, F9, and F7. From inspecting the direction of influence of the above-mentioned variables, they can be referred to as the positively contributing variables because they increase the response of the classifier, increasing the odds in favour of the assigned label, C1. Other positive variables that support the prediction of C1 are F2, F10, F6, and F8, however, unlike the top positive ones, these variables have only moderate control on the classifier. Just four of all the input variables are shown to reduce the probability that C1 is the correct label and these variables are F3, F1, F12, and F5 since their respective values cause the classification judgement to shift in the direction of C2. In summary, given that the confidence level in the C1 prediction is 97.82%, it is obvious that the negative contributions of F3, F1, F12, and F5 result in only a marginal decrease in the certainty or confidence level.

{'C1': '97.82%', 'C2': '2.18%'}

{'F4': 'Capital Gain', 'F11': 'Marital Status', 'F9': 'Relationship', 'F7': 'Age', 'F6': 'Education-Num', 'F8': 'Hours per week', 'F5': 'Occupation', 'F2': 'Capital Loss', 'F12': 'Sex', 'F10': 'Education', 'F1': 'Race', 'F14': 'fnlwgt', 'F3': 'Country', 'F13': 'Workclass'}

{'F11': 'F4', 'F6': 'F11', 'F8': 'F9', 'F1': 'F7', 'F5': 'F6', 'F13': 'F8', 'F7': 'F5', 'F12': 'F2', 'F10': 'F12', 'F4': 'F10', 'F9': 'F1', 'F3': 'F14', 'F14': 'F3', 'F2': 'F13'}

{'C1': 'C1', 'C2': 'C2'}

Under 50K

{'C1': 'Under 50K', 'C2': 'Above 50K'}

RandomForestClassifier

C2

Broadband Sevice Signup

The selected case is labelled as C2 with close to an 85.0% confidence level, hinting that there is a smaller chance that it could be C1. The most important variables when determining the label for this case are F15, F13, F42, and F21. The variables with moderate influence include F10, F25, F11, and F27. However, the last three ranked variables according to their respective impacts on the model for the case under consideration are F38, F41, and F35. Significantly increasing the odds of the predicted label are the variables F15 and F42. Conversely, the F13 has the strongest impact, driving the classification verdict towards C1. Other features with similar direction of influence as F13 are F17, F11, F30, F12, and F16.

{'C1': '15.00%', 'C2': '85.00%'}

{'F15': 'X32', 'F13': 'X38', 'F42': 'X35', 'F21': 'X24', 'F20': 'X31', 'F17': 'X19', 'F10': 'X42', 'F25': 'X12', 'F11': 'X21', 'F27': 'X7', 'F31': 'X22', 'F22': 'X27', 'F4': 'X11', 'F9': 'X41', 'F7': 'X1', 'F39': 'X16', 'F30': 'X33', 'F26': 'X6', 'F12': 'X10', 'F16': 'X4', 'F24': 'X5', 'F18': 'X37', 'F29': 'X39', 'F3': 'X40', 'F6': 'X36', 'F14': 'X34', 'F2': 'X2', 'F33': 'X30', 'F28': 'X28', 'F1': 'X26', 'F36': 'X25', 'F23': 'X3', 'F8': 'X23', 'F32': 'X20', 'F19': 'X18', 'F5': 'X17', 'F34': 'X15', 'F38': 'X14', 'F41': 'X13', 'F35': 'X9', 'F40': 'X8', 'F37': 'X29'}

{'F29': 'F15', 'F35': 'F13', 'F32': 'F42', 'F22': 'F21', 'F28': 'F20', 'F17': 'F17', 'F38': 'F10', 'F10': 'F25', 'F19': 'F11', 'F5': 'F27', 'F20': 'F31', 'F25': 'F22', 'F9': 'F4', 'F39': 'F9', 'F40': 'F7', 'F14': 'F39', 'F30': 'F30', 'F4': 'F26', 'F8': 'F12', 'F3': 'F16', 'F41': 'F24', 'F34': 'F18', 'F36': 'F29', 'F37': 'F3', 'F33': 'F6', 'F31': 'F14', 'F1': 'F2', 'F27': 'F33', 'F26': 'F28', 'F24': 'F1', 'F23': 'F36', 'F2': 'F23', 'F21': 'F8', 'F18': 'F32', 'F16': 'F19', 'F15': 'F5', 'F13': 'F34', 'F12': 'F38', 'F11': 'F41', 'F7': 'F35', 'F6': 'F40', 'F42': 'F37'}

{'C1': 'C1', 'C2': 'C2'}

Yes

{'C1': 'No', 'C2': 'Yes'}

LogisticRegression

C1

Australian Credit Approval

The probable label for the given case is C1 since its associated predicted probability is 91.85% compared to the 8.15% of C2. The input variables mostly responsible for the above prediction verdict are F11, F10, and F6, however, the values of F8, F14, and F3 are deemed less relevant by the model in this case. The attributions of the input variables can be either positive or negative, depending on the direction of influence on the model. Among the variables, the ones with negative attributions that decrease the probability that C1 is the correct label are F5, F4, F1, F8, F14, and F3. On the contrary, F11, F10, F6, F9, F7, F13, and F12 are some of the remaining variables that increase the likelihood of C1 being the correct label. Based on the attributions of the variables, we can conclude that the collective impact of the negative variables is not strong enough to shift the prediction verdict away from C1, resulting in only a marginal uncertainty in the assigned label.

{'C2': '8.15%', 'C1': '91.85%'}

{'F11': 'A8', 'F10': 'A14', 'F6': 'A9', 'F5': 'A13', 'F12': 'A5', 'F4': 'A11', 'F1': 'A12', 'F9': 'A7', 'F7': 'A4', 'F13': 'A10', 'F2': 'A6', 'F8': 'A1', 'F14': 'A2', 'F3': 'A3'}

{'F8': 'F11', 'F14': 'F10', 'F9': 'F6', 'F13': 'F5', 'F5': 'F12', 'F11': 'F4', 'F12': 'F1', 'F7': 'F9', 'F4': 'F7', 'F10': 'F13', 'F6': 'F2', 'F1': 'F8', 'F2': 'F14', 'F3': 'F3'}

{'C1': 'C2', 'C2': 'C1'}

Class 2

{'C2': 'Class 1', 'C1': 'Class 2'}

KNeighborsClassifier

C2

Tic-Tac-Toe Strategy

The classification verdict of the model for the case under consideration has a 50.10% chance of being C1. But based on the estimated likelihoods indicated above, it is possible to deduce that the model is extremely doubtful about the classification's validity. The following variables have the most attributions to the aforementioned prediction decisions: F1, F6, F2, and F9. F8 and F5 are the least important, whereas the values of the variables F4, F7, and F3 had only a moderate impact. Regarding the direction of influence of the variables, F1, F7, F8, and F5 are the ones driving the classification higher towards the C1 label and away from C2. However, factoring the likelihood of the C2 label, the negative variables, F6, F9, F2, F4, and F3, successfully cast doubt on the validity of the assigned label. In simple terms, the negative contributions from F6, F9, and F2 can easily explain the uncertainty associated with the class label assignment for the case under consideration here.

{'C1': '50.10%', 'C2': '49.90%'}

{'F1': 'middle-middle-square', 'F6': 'top-left-square', 'F9': 'bottom-left-square', 'F2': 'bottom-right-square', 'F4': 'top-middle-square', 'F7': ' top-right-square', 'F3': 'middle-right-square', 'F8': 'bottom-middle-square', 'F5': 'middle-left-square'}

{'F5': 'F1', 'F1': 'F6', 'F7': 'F9', 'F9': 'F2', 'F2': 'F4', 'F3': 'F7', 'F6': 'F3', 'F8': 'F8', 'F4': 'F5'}

{'C2': 'C1', 'C1': 'C2'}

player B lose

{'C1': 'player B win', 'C2': 'player B lose'}

LogisticRegression

C2

Used Cars Price-Range Prediction

The algorithm classified the given data as C2 with close to 99.32% certainty since the prediction likelihood of C1 is only 0.68%. The abovementioned prediction verdict is largely due to the influence of F10, F2, and F1 while the other influential features include F7, F3, and F4. However, F5, F8, F9, and F6 are shown to have smaller contributions to the decision made here. Not all the features have positive contributions, and F1, F3, and F4 are known as negative features since for the given case, they reduce the likelihood of the assigned label and hence they favour or support labelling the case as C1 instead.

{'C1': '0.68%', 'C2': '99.32%'}

{'F10': 'Power', 'F2': 'car_age', 'F1': 'Name', 'F7': 'Fuel_Type', 'F3': 'Seats', 'F4': 'Transmission', 'F5': 'Mileage', 'F8': 'Owner_Type', 'F9': 'Kilometers_Driven', 'F6': 'Engine'}

{'F4': 'F10', 'F5': 'F2', 'F6': 'F1', 'F7': 'F7', 'F10': 'F3', 'F8': 'F4', 'F2': 'F5', 'F9': 'F8', 'F1': 'F9', 'F3': 'F6'}

{'C1': 'C1', 'C2': 'C2'}

High

{'C1': 'Low', 'C2': 'High'}

LogisticRegression

C1

Flight Price-Range Classification

The model is quite certain that C1 is the most likely class for the current scenario. C1 has a 90.48% chance of being correct, implying that any of the other labels is highly unlikely. F10 and F4 are the most relevant variables influencing the abovementioned classification decision but all other factors or variables are proven to have a moderate or minor influence. Fortunately, the top variables, F4 and F10, have an impact on the model that is positive, boosting the chance of C1. Furthermore, whereas F1 and F5 force the model to forecast C1, the variables F8, F11, F9, and F6 are forcing the model to assign a different label. Finally, several variables have a very minor influence on the model's final forecast here, but F12, F9, and F6 are shown to have the least contributions.

{'C1': '90.48%', 'C3': '9.51%', 'C2': '0.01%'}

{'F4': 'Total_Stops', 'F10': 'Airline', 'F5': 'Destination', 'F1': 'Arrival_hour', 'F8': 'Source', 'F7': 'Duration_hours', 'F2': 'Dep_hour', 'F11': 'Dep_minute', 'F3': 'Arrival_minute', 'F12': 'Journey_month', 'F9': 'Journey_day', 'F6': 'Duration_mins'}

{'F12': 'F4', 'F9': 'F10', 'F11': 'F5', 'F5': 'F1', 'F10': 'F8', 'F7': 'F7', 'F3': 'F2', 'F4': 'F11', 'F6': 'F3', 'F2': 'F12', 'F1': 'F9', 'F8': 'F6'}

{'C1': 'C1', 'C2': 'C3', 'C3': 'C2'}

Low

{'C1': 'Low', 'C3': 'Moderate', 'C2': 'High'}

LogisticRegression

C2

Suspicious Bidding Identification

The label assigned by the model is C2 with a higher predicted confidence level of 99.99%, meaning the probability of C1 being the correct label is virtually equal to zero. The classification decision above is mainly due to the influence of the features F5, F3, F8, and F4, however, the remaining features have very marginal contributions to the decision. Among the features, only F1 and F6 are shown to have a negative impact, reducing the likelihood of the assigned label. However, this negative influence is very weak compared to that of the top positive features, F5, F3, F8, and F4.

{'C2': '99.99%', 'C1': '0.01%'}

{'F5': 'Z3', 'F3': 'Z8', 'F8': 'Z4', 'F4': 'Z2', 'F9': 'Z5', 'F7': 'Z7', 'F1': 'Z1', 'F2': 'Z6', 'F6': 'Z9'}

{'F3': 'F5', 'F8': 'F3', 'F4': 'F8', 'F2': 'F4', 'F5': 'F9', 'F7': 'F7', 'F1': 'F1', 'F6': 'F2', 'F9': 'F6'}

{'C1': 'C2', 'C2': 'C1'}

Normal

{'C2': 'Normal', 'C1': 'Suspicious'}

SVC

C1

German Credit Evaluation

With respect to the given case, the classification algorithm employed here generates C1 as the most probable class since the probability of C2 is 41.63% while that of C1 is 58.37%. F1, F7, and F4 are the most influential features resulting in the classification decision mentioned above, whereas the least relevant features are F2 and F5. As indicated by the prediction probabilities across the classes, the confidence in the labelling decision here is not perfect, which can be attributed to the influence of the negative features F1, F7, F4, and F8. On the other hand, the moderate positive influence of F3, F6, F9, F2, and F5 explains the algorithm's decision to label the case as C1 with such an average level of confidence.

{'C1': '58.37%', 'C2': '41.63%'}

{'F1': 'Checking account', 'F7': 'Duration', 'F4': 'Saving accounts', 'F6': 'Sex', 'F3': 'Purpose', 'F9': 'Age', 'F8': 'Housing', 'F2': 'Job', 'F5': 'Credit amount'}

{'F6': 'F1', 'F8': 'F7', 'F5': 'F4', 'F2': 'F6', 'F9': 'F3', 'F1': 'F9', 'F4': 'F8', 'F3': 'F2', 'F7': 'F5'}

{'C1': 'C1', 'C2': 'C2'}

Good Credit

{'C1': 'Good Credit', 'C2': 'Bad Credit'}

GradientBoostingClassifier

C1

German Credit Evaluation

The most likely label for the provided data instance, according to the predictive algorithm used here, is C1. The confidence level associated with the above prediction decision is 64.62 percent, which means C2 has a 35.38 percent chance of being correct. The following input features can be prioritised in decreasing order according to their relative degrees of influence: F3, F7, F2, F6, F4, F1, F8, F9, and F5. As a result, the algorithm places little emphasis or attention on the values of F7 and F3 when classifying the given case whilst the values of F9, F8, and F5 are the most relevant. Regarding the direction of influence or impact of the input features, F5, F1, F8, and F2 are considered negative features because their contributions reduce the likelihood of C1 being the correct label. Positive features such as F4, F6, and F9, however, push the algorithm closer to assigning C1 to the situation in question.

{'C1': '64.62%', 'C2': '35.38%'}

{'F5': 'Saving accounts', 'F9': 'Sex', 'F8': 'Duration', 'F1': 'Housing', 'F4': 'Checking account', 'F6': 'Purpose', 'F2': 'Credit amount', 'F7': 'Age', 'F3': 'Job'}

{'F5': 'F5', 'F2': 'F9', 'F8': 'F8', 'F4': 'F1', 'F6': 'F4', 'F9': 'F6', 'F7': 'F2', 'F1': 'F7', 'F3': 'F3'}

{'C1': 'C1', 'C2': 'C2'}

Good Credit

{'C1': 'Good Credit', 'C2': 'Bad Credit'}

SVC

C1

Real Estate Investment

C1 is the label predicted by the classifier for the case or example under consideration the confidence in the above prediction is about 96.35%. It is important to take into consideration, however, that there is also a very small chance equal to 3.65% that the correct label could be C2. The ranking of the features according to their respective contributions to the decision above is as follows: The top features with significant influences are F9, F19, and F1. The remaining features with moderate contributions are: F8, F15, F13, F18, F4, F6, F12, F20, F7, F10, F5, F2, and F16. Finally, the values of F14, F17, F11, and F3 are shown to have a very low impact on the prediction of C1 for the case under consideration. The assessment below only considers the features shown to have the most relevant impact in terms of the direction of the prediction here. Among the most contributing features, only F8 and F15 have a negative influence, while the remaining ones, F9, F19, F1, and F13, are shown to have positive contributions to the prediction for the case. Looking at the cumulative influences of each set of positive and negative features, it is not strange that the label assigned is C1 with a confidence level of 96.35%.

{'C2': '3.65%', 'C1': '96.35%'}

{'F9': 'Feature7', 'F19': 'Feature4', 'F1': 'Feature14', 'F8': 'Feature2', 'F15': 'Feature8', 'F13': 'Feature1', 'F18': 'Feature13', 'F4': 'Feature6', 'F6': 'Feature10', 'F12': 'Feature15', 'F20': 'Feature18', 'F7': 'Feature9', 'F10': 'Feature12', 'F5': 'Feature16', 'F2': 'Feature19', 'F16': 'Feature5', 'F14': 'Feature11', 'F17': 'Feature20', 'F11': 'Feature3', 'F3': 'Feature17'}

{'F11': 'F9', 'F9': 'F19', 'F17': 'F1', 'F1': 'F8', 'F3': 'F15', 'F7': 'F13', 'F16': 'F18', 'F10': 'F4', 'F13': 'F6', 'F4': 'F12', 'F19': 'F20', 'F12': 'F7', 'F15': 'F10', 'F18': 'F5', 'F5': 'F2', 'F2': 'F16', 'F14': 'F14', 'F20': 'F17', 'F8': 'F11', 'F6': 'F3'}

{'C1': 'C2', 'C2': 'C1'}

Invest

{'C2': 'Ignore', 'C1': 'Invest'}

DecisionTreeClassifier

C1

Insurance Churn

The model predicted the C1 class with very high confidence of 93.27%, hence we can conclude that there is only a 6.73% chance that the true label is C2. Two features have a very strong positive influence on the prediction of the C1 class and they are F7 and F3. The following features have a medium impact and are listed in decreasing order of influence: F5 and F12 have a negative influence, while F11 and F1 have a positive influence on the prediction of C1. F1, F8, and F14 have a positive influence on the prediction of the C1 class, while F4, F9, F13, and F2 influence the prediction negatively. Those with the least contribution regarding the model's decision for this case are shown to be F10, F16, F14, and F6. Among these least contributing features F10 and F6 are shown to have negative contributions whereas F14 and F16 contribute positively in favour of the assigned label.

{'C2': '6.73%', 'C1': '93.27%'}

{'F7': 'feature15', 'F3': 'feature14', 'F5': 'feature10', 'F12': 'feature11', 'F11': 'feature5', 'F1': 'feature13', 'F4': 'feature4', 'F2': 'feature3', 'F9': 'feature12', 'F13': 'feature1', 'F8': 'feature7', 'F15': 'feature2', 'F14': 'feature6', 'F10': 'feature0', 'F16': 'feature9', 'F6': 'feature8'}

{'F9': 'F7', 'F8': 'F3', 'F4': 'F5', 'F5': 'F12', 'F15': 'F11', 'F7': 'F1', 'F14': 'F4', 'F13': 'F2', 'F6': 'F9', 'F11': 'F13', 'F1': 'F8', 'F12': 'F15', 'F16': 'F14', 'F10': 'F10', 'F3': 'F16', 'F2': 'F6'}

{'C1': 'C2', 'C2': 'C1'}

Leave

{'C2': 'Stay', 'C1': 'Leave'}

BernoulliNB

C1

Customer Churn Modelling

This case or instance is labelled as C1 with a very high confidence level, however, the classifier estimates that C2 could be the correct label with a prediction likelihood of about 5.75%. The values F9, F3, and F1 played a major role in the aforementioned labelling choice and because F4 and F8 have minimal attributions, they are the lowest rated features. F3 and F9 have values, which increases the probability that C1 is the correct label. Other variables that drive the clasifier towards assigning the predicted class are F2 and F6. Here, F3, F6, F9, and F2, are referred to as positive input variables since their contributions are towards the generated C1 label. In contrast, the remaining six variables are shown to have a negative influence on the classifier, indicating that the correct label could be C2 instead of the C1 selected by the classifier and the strongest negative variables are F1, F7, and F10.

{'C1': '94.25%', 'C2': '5.75%'}

{'F9': 'IsActiveMember', 'F3': 'NumOfProducts', 'F1': 'Gender', 'F7': 'Geography', 'F10': 'Age', 'F5': 'CreditScore', 'F2': 'EstimatedSalary', 'F6': 'Balance', 'F4': 'HasCrCard', 'F8': 'Tenure'}

{'F9': 'F9', 'F7': 'F3', 'F3': 'F1', 'F2': 'F7', 'F4': 'F10', 'F1': 'F5', 'F10': 'F2', 'F6': 'F6', 'F8': 'F4', 'F5': 'F8'}

{'C1': 'C1', 'C2': 'C2'}

Stay

{'C1': 'Stay', 'C2': 'Leave'}

LogisticRegression

C1

Printer Sales

The prediction likelihood of class C1 is 73.85%, making it the most probable label for the given case. When making the above prediction, the input features are shown to have some degree of influence on the decision made by the classifier. While features such as F16, F25, and F19 have very low contributions to the classification, the features F22 and F26 are shown to be the main contributors to the decision. Finally, the features with moderate contributions are 21, F11, 42, F18, F12, and F10. As indicated by the prediction likelihoods across the classes, the classifier is shown to have a little doubt in the correctness or validity of C1, and the main features resulting in this little uncertainty are the negative features F22, F13, F10, F12, F5, F6, and F24. However, the values of F26, F11, F14, F18, F4, and F15 suggest that C1 is very likely the true label.

{'C1': '73.85%', 'C2': '26.15%'}

{'F26': 'X8', 'F22': 'X1', 'F11': 'X24', 'F14': 'X21', 'F13': 'X4', 'F10': 'X18', 'F18': 'X17', 'F12': 'X25', 'F15': 'X7', 'F5': 'X20', 'F6': 'X23', 'F4': 'X9', 'F21': 'X2', 'F7': 'X22', 'F3': 'X16', 'F17': 'X10', 'F1': 'X15', 'F23': 'X14', 'F24': 'X26', 'F9': 'X19', 'F16': 'X13', 'F25': 'X12', 'F19': 'X11', 'F2': 'X6', 'F8': 'X5', 'F20': 'X3'}

{'F8': 'F26', 'F1': 'F22', 'F24': 'F11', 'F21': 'F14', 'F4': 'F13', 'F18': 'F10', 'F17': 'F18', 'F25': 'F12', 'F7': 'F15', 'F20': 'F5', 'F23': 'F6', 'F9': 'F4', 'F2': 'F21', 'F22': 'F7', 'F16': 'F3', 'F10': 'F17', 'F15': 'F1', 'F14': 'F23', 'F26': 'F24', 'F19': 'F9', 'F13': 'F16', 'F12': 'F25', 'F11': 'F19', 'F6': 'F2', 'F5': 'F8', 'F3': 'F20'}

{'C1': 'C1', 'C2': 'C2'}

Less

{'C1': 'Less', 'C2': 'More'}

LogisticRegression

C1

Printer Sales

According to the model, the probability of C2 is 12.35% and that of C1 is 87.65% meaning C1 is the most probable label for the given case. The variables with the majority influence on the abovementioned decision are F25, F3, F12, F8, F20, and F21 whereas variables F16, F23, F4, F9, F6, and F26 are shown to have little to no influence on the model's decision with respect to the given case. The contributions and influence of variables such as F19, F1, F24, and F17 can be described as moderate. Among the variables controlling the prediction decision here, F25, F20, F1, F17, F15, F7, F22, and F14 are the negative variables decreasing the model's response to the output of the label C1. Conversely, the highly influential variables F3, F12, F8, and F21 are the main drivers that contribute positively, increasing the probability of C1 being the correct class label. Overall, given that the variable with the highest influence on the model is F25, a negative variable, it is not unexpected that there is a little doubt in the classification decision here, as shown by the prediction probabilities across the classes.

{'C2': '12.35%', 'C1': '87.65%'}

{'F25': 'X8', 'F3': 'X21', 'F12': 'X1', 'F8': 'X25', 'F20': 'X2', 'F21': 'X24', 'F19': 'X15', 'F1': 'X4', 'F18': 'X20', 'F24': 'X10', 'F17': 'X5', 'F11': 'X6', 'F15': 'X11', 'F7': 'X12', 'F22': 'X9', 'F10': 'X26', 'F2': 'X23', 'F13': 'X7', 'F5': 'X14', 'F14': 'X17', 'F16': 'X18', 'F23': 'X19', 'F4': 'X16', 'F9': 'X13', 'F6': 'X22', 'F26': 'X3'}

{'F8': 'F25', 'F21': 'F3', 'F1': 'F12', 'F25': 'F8', 'F2': 'F20', 'F24': 'F21', 'F15': 'F19', 'F4': 'F1', 'F20': 'F18', 'F10': 'F24', 'F5': 'F17', 'F6': 'F11', 'F11': 'F15', 'F12': 'F7', 'F9': 'F22', 'F26': 'F10', 'F23': 'F2', 'F7': 'F13', 'F14': 'F5', 'F17': 'F14', 'F18': 'F16', 'F19': 'F23', 'F16': 'F4', 'F13': 'F9', 'F22': 'F6', 'F3': 'F26'}

{'C1': 'C2', 'C2': 'C1'}

More

{'C2': 'Less', 'C1': 'More'}

RandomForestClassifier

C1

Cab Surge Pricing System

With a moderately high level of confidence, C1 is assigned to the given case by the classifier and this is due to the fact that the other classes, C3 and C2, have likelihoods of 3.0% and 14.0%, respectively. Across the input features, only F11, F3, F9, and F12 are shown to contribute negatively, shifting the classification away from C1 and towards C3 and C2. On the contrary, the features such as F5, F6, F4, and F2 are among the positive set of features that drive the verdict in support of assigning C1 to the given case. From the attributions of the different features, F5 is the most relevant contributor to the classification made here, while F8, F10, and F7 are ranked as the least influential features and considering the direction of influence of each input feature, it is understandable why the classifier is certain about the decision made.

{'C3': '3.00%', 'C1': '83.00%', 'C2': '14.00%'}

{'F5': 'Type_of_Cab', 'F11': 'Destination_Type', 'F3': 'Trip_Distance', 'F9': 'Cancellation_Last_1Month', 'F6': 'Confidence_Life_Style_Index', 'F4': 'Var3', 'F12': 'Customer_Since_Months', 'F2': 'Life_Style_Index', 'F1': 'Var2', 'F8': 'Gender', 'F10': 'Var1', 'F7': 'Customer_Rating'}

{'F2': 'F5', 'F6': 'F11', 'F1': 'F3', 'F8': 'F9', 'F5': 'F6', 'F11': 'F4', 'F3': 'F12', 'F4': 'F2', 'F10': 'F1', 'F12': 'F8', 'F9': 'F10', 'F7': 'F7'}

{'C1': 'C3', 'C2': 'C1', 'C3': 'C2'}

C2

{'C3': 'Low', 'C1': 'Medium', 'C2': 'High'}

MLPClassifier

C1

Ethereum Fraud Detection

Considering the values of the input variables, the classification model is very confident that the most probable label is not C2 but C1. The top input variables receiving much consideration from the model to arrive at the classification verdict are F38, F28, F31, F15, and F22. Among these most influential variables, F38 and F28 are regarded as negatives since their contributions serve to swing the classification decision in the opposite direction. On the contrary, F31, F15, and F22 have a positive influence, increasing the model's response to favour labelling the given case as C1. Other positive variables include F17, F14, and F13, whereas the other negative ones include F26, F9, and F6. Input variables such as F7, F29, F21, and F36 are shown to have zero attributions, that is, their values are not paid enough attention to influence the model's decision with respect to the given case.

{'C1': '100.00%', 'C2': '0.00%'}

{'F38': 'Unique Received From Addresses', 'F28': ' ERC20 total Ether sent contract', 'F31': 'total ether received', 'F15': 'Number of Created Contracts', 'F22': 'Sent tnx', 'F26': ' ERC20 uniq rec token name', 'F9': ' ERC20 uniq rec contract addr', 'F14': 'max value received ', 'F17': 'total transactions (including tnx to create contract', 'F6': ' ERC20 uniq sent addr.1', 'F13': ' ERC20 uniq sent addr', 'F27': 'Received Tnx', 'F4': ' ERC20 uniq rec addr', 'F16': 'avg val sent', 'F5': 'min value received', 'F3': 'Unique Sent To Addresses', 'F33': ' ERC20 uniq sent token name', 'F37': ' Total ERC20 tnxs', 'F30': 'Time Diff between first and last (Mins)', 'F12': 'Avg min between received tnx', 'F7': 'total Ether sent', 'F29': 'min val sent', 'F21': 'avg val received', 'F36': ' ERC20 avg val sent', 'F34': ' ERC20 max val sent', 'F2': ' ERC20 min val sent', 'F23': ' ERC20 avg val rec', 'F10': ' ERC20 max val rec', 'F20': ' ERC20 min val rec', 'F24': 'max val sent', 'F25': 'min value sent to contract', 'F8': 'max val sent to contract', 'F18': ' ERC20 total ether sent', 'F19': ' ERC20 total Ether received', 'F32': 'avg value sent to contract', 'F11': 'total ether balance', 'F35': 'total ether sent contracts', 'F1': 'Avg min between sent tnx'}

{'F7': 'F38', 'F26': 'F28', 'F20': 'F31', 'F6': 'F15', 'F4': 'F22', 'F38': 'F26', 'F30': 'F9', 'F10': 'F14', 'F18': 'F17', 'F29': 'F6', 'F27': 'F13', 'F5': 'F27', 'F28': 'F4', 'F14': 'F16', 'F9': 'F5', 'F8': 'F3', 'F37': 'F33', 'F23': 'F37', 'F3': 'F30', 'F2': 'F12', 'F19': 'F7', 'F12': 'F29', 'F11': 'F21', 'F36': 'F36', 'F35': 'F34', 'F34': 'F2', 'F33': 'F23', 'F32': 'F10', 'F31': 'F20', 'F13': 'F24', 'F15': 'F25', 'F16': 'F8', 'F25': 'F18', 'F24': 'F19', 'F17': 'F32', 'F22': 'F11', 'F21': 'F35', 'F1': 'F1'}

{'C1': 'C1', 'C2': 'C2'}

Not Fraud

{'C1': 'Not Fraud', 'C2': 'Fraud'}

SVC

C1

Job Change of Data Scientists

The odds are in favour of label C1 given that the probability of it being the correct label for the case under consideration is 81.32%. However, the likelihood of label C2 is 18.68%. The classification decision above is mainly due to the values of F8, F12, F9, and F5. The feature with the least attribution to the model's output label here is F6. The features F8, F9, and F12 have very strong positive contributions to the prediction, increasing the odds of the label C1. Other features with positive attribution in support of C1 are F7, F4, and F11. Unlike the features stated above, the remaining features, F5, F10, F2, F1, F3, and F6, have values that shift the final prediction verdict in the direction of C2 and account for its 18.68% likelihood.

{'C1': '81.32%', 'C2': '18.68%'}

{'F8': 'city', 'F12': 'company_type', 'F9': 'city_development_index', 'F5': 'education_level', 'F10': 'enrolled_university', 'F2': 'gender', 'F7': 'relevent_experience', 'F1': 'training_hours', 'F3': 'major_discipline', 'F4': 'company_size', 'F11': 'experience', 'F6': 'last_new_job'}

{'F3': 'F8', 'F11': 'F12', 'F1': 'F9', 'F7': 'F5', 'F6': 'F10', 'F4': 'F2', 'F5': 'F7', 'F2': 'F1', 'F8': 'F3', 'F10': 'F4', 'F9': 'F11', 'F12': 'F6'}

{'C1': 'C1', 'C2': 'C2'}

Stay

{'C1': 'Stay', 'C2': 'Leave'}

DNN

C2

Concrete Strength Classification

The model predicted C2 with a high probability equal to 88.70%, whereas C1 has only a 11.30% likelihood of being the true label. Considering the predicted likelihood of C1, there is only little confidence in its correctness as the true label for the case here. The value of F4 has a large negative influence on the C2 classification decision, while F8 is the top positive feature. F1, F5, F3, and F2 all have positive impacts on the C2 prediction, with F1 and F8 having the highest influence, F3 and F2 having low influence, and F5 being somewhere in the middle. Broadly speaking, the negative influences of F4, F6, and F7 only succeed in driving the decision slightly away from C2 towards the other label as shown by the predicted probabilities across the classes.

{'C2': '88.70%', 'C1': '11.30%'}

{'F4': 'coarseaggregate', 'F8': 'age_days', 'F1': 'superplasticizer', 'F5': 'cement', 'F3': 'water', 'F2': 'fineaggregate', 'F7': 'slag', 'F6': 'flyash'}

{'F6': 'F4', 'F8': 'F8', 'F5': 'F1', 'F1': 'F5', 'F4': 'F3', 'F7': 'F2', 'F2': 'F7', 'F3': 'F6'}

{'C1': 'C2', 'C2': 'C1'}

Weak

{'C2': 'Weak', 'C1': 'Strong'}

BernoulliNB

C2

Suspicious Bidding Identification

The algorithm's predicted output label for the given case is C2 with a very strong confidence level equal to 100.0%; hence C1 can't be the true label. Among the features, the most relevant ones are F3, F6, and F8 with very significant impact, pushing the prediction decision away from C1 towards C2. The next set of attributes, F9, F2, and F5, offer a moderate shift towards C2 coupled with marginal positive contribution from F4 and F1. From the above statements, all the features are shown to support the label assignment decision in the case under consideration. Consequently, it is no wonder that the algorithm has 100.0% confidence in the output decision or verdict above.

{'C2': '100.00%', 'C1': '0.00%'}

{'F3': 'Z3', 'F6': 'Z8', 'F8': 'Z2', 'F9': 'Z7', 'F7': 'Z5', 'F2': 'Z4', 'F5': 'Z6', 'F4': 'Z1', 'F1': 'Z9'}

{'F3': 'F3', 'F8': 'F6', 'F2': 'F8', 'F7': 'F9', 'F5': 'F7', 'F4': 'F2', 'F6': 'F5', 'F1': 'F4', 'F9': 'F1'}

{'C1': 'C2', 'C2': 'C1'}

Normal

{'C2': 'Normal', 'C1': 'Suspicious'}

RandomForestClassifier

C2

Health Care Services Satisfaction Prediction

In this case, the prediction algorithm is not 100.0% certain that the correct label for the given case is C2, since there is a 43.49% chance that the right label could be C1 instead. The algorithm's decision to label the case as C2 mainly stems from the influence of features such as F1, F12, F15, F3, and F14. On the other hand, little consideration is paid to the values of the least ranked features, F9, F6, and F7. Within the top-ranked features, F14 and F3 have a negative impact, increasing the prediction probability of label C1. Further decreasing the likelihood of the C2 class are the negative features are F2, F11, F16, and F7. However, all the remaining features strongly or moderately push for the classification output to be C2 and the notable positive features are F1, F12, and F15. Considering all the features' attributions, the uncertainty or doubt in the classification could be attributed to the algorithm's paying too much attention to the values of the negative features.

{'C1': '43.49%', 'C2': '56.51%'}

{'F1': 'Quality\\/experience dr.', 'F12': 'Exact diagnosis', 'F15': 'Hygiene and cleaning', 'F14': 'Specialists avaliable', 'F3': 'Modern equipment', 'F4': 'hospital rooms quality', 'F13': 'Admin procedures', 'F8': 'avaliablity of drugs', 'F2': 'parking, playing rooms, caffes', 'F11': 'Time waiting', 'F16': 'friendly health care workers', 'F5': 'Communication with dr', 'F10': 'waiting rooms', 'F9': 'Check up appointment', 'F6': 'lab services', 'F7': 'Time of appointment'}

{'F6': 'F1', 'F9': 'F12', 'F4': 'F15', 'F7': 'F14', 'F10': 'F3', 'F15': 'F4', 'F3': 'F13', 'F13': 'F8', 'F16': 'F2', 'F2': 'F11', 'F11': 'F16', 'F8': 'F5', 'F14': 'F10', 'F1': 'F9', 'F12': 'F6', 'F5': 'F7'}

{'C1': 'C1', 'C2': 'C2'}

Satisfied

{'C1': 'Dissatisfied', 'C2': 'Satisfied'}

LogisticRegression

C1

Flight Price-Range Classification

Mainly based on the information on the case given, the classifier's output decision is as follows: C1 is the most probable label, followed by C3 and C2, with C2 being the least. To be specific, the prediction probabilities across the classes are as follows: C2 has 4.34%, C3 has 21.64%, and C1 has 74.0% chance of being the true label. The moderately high classification confidence is largely due to the impact of F3, F1, and F9. However, the values of F4 and F2 received very little consideration when the classifier was picking the most probable label for the given case. With respect to the direction of influence of the features, F3, F1, F10, F6, and F2 have varying degrees of positive contributions, driving the classifier to label the case as C1. On the contrary, F9, F12, F8, and F11 are among the negative features, shifting the classification decision in a direction away from C1.

{'C2': '4.34%', 'C3': '21.64%', 'C1': '74.02%'}

{'F3': 'Airline', 'F1': 'Total_Stops', 'F9': 'Source', 'F12': 'Arrival_minute', 'F8': 'Arrival_hour', 'F11': 'Dep_minute', 'F5': 'Duration_hours', 'F7': 'Journey_month', 'F10': 'Journey_day', 'F6': 'Duration_mins', 'F4': 'Destination', 'F2': 'Dep_hour'}

{'F9': 'F3', 'F12': 'F1', 'F10': 'F9', 'F6': 'F12', 'F5': 'F8', 'F4': 'F11', 'F7': 'F5', 'F2': 'F7', 'F1': 'F10', 'F8': 'F6', 'F11': 'F4', 'F3': 'F2'}

{'C1': 'C2', 'C2': 'C3', 'C3': 'C1'}

High

{'C2': 'Low', 'C3': 'Moderate', 'C1': 'High'}

LogisticRegression

C1

Used Cars Price-Range Prediction

The model predicted C1 for the case under consideration which a predicted likelihood of 67.95% whereas, that of C2 is 32.05%. The top influencing features ordered from highest to lowest, are F9, F8, F6 and F1, and among them only F6 is shown to have positive attribution in support of the model's decision. F5, F10, and F3 have a smaller positive influence on the prediction, while F4 has an even smaller negative impact. F7 is the least relevant feature, and hence its negative attribution has very little influence on the model for this case.

{'C2': '32.05%', 'C1': '67.95%'}

{'F9': 'Fuel_Type', 'F8': 'Seats', 'F6': 'car_age', 'F1': 'Name', 'F2': 'Owner_Type', 'F5': 'Power', 'F3': 'Engine', 'F4': 'Transmission', 'F10': 'Mileage', 'F7': 'Kilometers_Driven'}

{'F7': 'F9', 'F10': 'F8', 'F5': 'F6', 'F6': 'F1', 'F9': 'F2', 'F4': 'F5', 'F3': 'F3', 'F8': 'F4', 'F2': 'F10', 'F1': 'F7'}

{'C1': 'C2', 'C2': 'C1'}

High

{'C2': 'Low', 'C1': 'High'}

LogisticRegression

C1

Vehicle Insurance Claims

The case under consideration is labelled as C1 by the model employed for this classification problem. However, according to the model, there is a 45.34% chance that C2 could be the label, presenting some level of uncertainty in the classification verdict made here. F7, F15, F8, F16, F2, and F1 are the top features identified as very important to the model's decision, whereas those with negligible contributions include F33, F10, F25, F13, and F4. Across the input features, those with a negative influence that motivates the classification output to be C2 are mainly F15, F5, F16, and F21. The other negative features contributing to the predicted likelihood of C2 are F9, F17, F26, and F20. Contradicting all the negative features and motivating the prediction output to be C1 are the positive features F7, F8, F1, F2, F28, F19, and F23. In conclusion, it is very surprising to see that the confidence level of the C1 prediction is only 54.66% given the very strong positive contribution of F7, but one can say that the negative features successfully cast doubt on the decision with regards to the case under consideration.

{'C2': '45.34%', 'C1': '54.66%'}

{'F7': 'insured_hobbies', 'F15': 'incident_severity', 'F8': 'auto_make', 'F16': 'number_of_vehicles_involved', 'F1': 'insured_education_level', 'F2': 'collision_type', 'F5': 'insured_occupation', 'F21': 'incident_city', 'F28': 'incident_type', 'F23': 'auto_year', 'F9': 'insured_relationship', 'F20': 'authorities_contacted', 'F17': 'witnesses', 'F26': 'bodily_injuries', 'F19': 'umbrella_limit', 'F30': 'policy_csl', 'F24': 'insured_sex', 'F3': 'injury_claim', 'F27': 'policy_deductable', 'F14': 'total_claim_amount', 'F10': 'police_report_available', 'F33': 'property_damage', 'F25': 'incident_state', 'F13': 'policy_annual_premium', 'F4': 'capital-loss', 'F11': 'insured_zip', 'F18': 'capital-gains', 'F12': 'incident_hour_of_the_day', 'F29': 'policy_state', 'F22': 'age', 'F31': 'vehicle_claim', 'F6': 'property_claim', 'F32': 'months_as_customer'}

{'F23': 'F7', 'F27': 'F15', 'F33': 'F8', 'F10': 'F16', 'F21': 'F1', 'F26': 'F2', 'F22': 'F5', 'F30': 'F21', 'F25': 'F28', 'F17': 'F23', 'F24': 'F9', 'F28': 'F20', 'F12': 'F17', 'F11': 'F26', 'F5': 'F19', 'F19': 'F30', 'F20': 'F24', 'F14': 'F3', 'F3': 'F27', 'F13': 'F14', 'F32': 'F10', 'F31': 'F33', 'F29': 'F25', 'F4': 'F13', 'F8': 'F4', 'F6': 'F11', 'F7': 'F18', 'F9': 'F12', 'F18': 'F29', 'F2': 'F22', 'F16': 'F31', 'F15': 'F6', 'F1': 'F32'}

{'C1': 'C2', 'C2': 'C1'}

Fraud

{'C2': 'Not Fraud', 'C1': 'Fraud'}

GradientBoostingClassifier

C2

Food Ordering Customer Churn Prediction

With a certainty level of 82.07 percent, the label choice for the given case is C2 and in a nutshell, the likelihood of C1 having the correct label is only 17.93%. The contributions of features like F30, F33, F38, and F36 are largely responsible for the classification above. The following three, with modest impact, are F23, F28, and F5. However, while choosing the proper label for a given case, the classifier does not consider all of the features. F39, F35, F20, and F14 are notable but insignificant features. F30, F33, F38, and F36 are the top features, with considerable positive contributions supporting the assignment of label C2. F23, F28, F10, and F40 are the top negative features that cause the classification to swing in a different direction. To bring things into perspective, due to the fact that the bulk of important attributes have positive attributions, it's not surprising that the classifier is confident that C2 rather than C1 is the correct label.

{'C1': '17.93%', 'C2': '82.07%'}

{'F30': 'More restaurant choices', 'F33': 'Ease and convenient', 'F38': 'Bad past experience', 'F36': 'Time saving', 'F23': 'Easy Payment option', 'F28': 'Good Tracking system', 'F5': 'Wrong order delivered', 'F10': 'Influence of rating', 'F12': 'Late Delivery', 'F40': 'Less Delivery time', 'F15': 'Long delivery time', 'F45': 'Delivery person ability', 'F43': 'Order placed by mistake', 'F46': 'More Offers and Discount', 'F26': 'Freshness ', 'F1': 'Unavailability', 'F19': 'Delay of delivery person picking up food', 'F11': 'Poor Hygiene', 'F13': 'Order Time', 'F22': 'Delay of delivery person getting assigned', 'F39': 'High Quality of package', 'F35': 'Residence in busy location', 'F20': 'Good Taste ', 'F14': 'Temperature', 'F34': 'Google Maps Accuracy', 'F7': 'Good Road Condition', 'F2': 'Number of calls', 'F29': 'Low quantity low time', 'F42': 'Politeness', 'F6': 'Maximum wait time', 'F4': 'Age', 'F25': 'Influence of time', 'F16': 'Missing item', 'F21': 'Family size', 'F44': 'Unaffordable', 'F32': 'Health Concern', 'F3': 'Self Cooking', 'F18': 'Good Food quality', 'F17': 'Perference(P2)', 'F9': 'Perference(P1)', 'F8': 'Educational Qualifications', 'F31': 'Monthly Income', 'F24': 'Occupation', 'F37': 'Marital Status', 'F41': 'Gender', 'F27': 'Good Quantity'}

{'F12': 'F30', 'F10': 'F33', 'F21': 'F38', 'F11': 'F36', 'F13': 'F23', 'F16': 'F28', 'F27': 'F5', 'F38': 'F10', 'F19': 'F12', 'F39': 'F40', 'F24': 'F15', 'F37': 'F45', 'F29': 'F43', 'F14': 'F46', 'F43': 'F26', 'F22': 'F1', 'F26': 'F19', 'F20': 'F11', 'F31': 'F13', 'F25': 'F22', 'F40': 'F39', 'F33': 'F35', 'F45': 'F20', 'F44': 'F14', 'F34': 'F34', 'F35': 'F7', 'F41': 'F2', 'F36': 'F29', 'F42': 'F42', 'F32': 'F6', 'F1': 'F4', 'F30': 'F25', 'F28': 'F16', 'F7': 'F21', 'F23': 'F44', 'F18': 'F32', 'F17': 'F3', 'F15': 'F18', 'F9': 'F17', 'F8': 'F9', 'F6': 'F8', 'F5': 'F31', 'F4': 'F24', 'F3': 'F37', 'F2': 'F41', 'F46': 'F27'}

{'C1': 'C1', 'C2': 'C2'}

Go Away

{'C1': 'Return', 'C2': 'Go Away'}

DNN

C2

Concrete Strength Classification

C2 was predicted with a high degree of certainty by the model since the likelihood of the alternative class is only 11.30%. The value of F8 has a significant negative impact on the classification choice, whereas F6 has a moderately positive contribution. F4, F3, F7, and F1 all have a favourable or positive impact on the C2 prediction, with F1 having the most positive impact, F3 and F4 having the least, and F7 being in between. F2 and F5 have a minor negative influence on the class assignment here, which together with F8 contributing to the decrease in the liklihood of C2.

{'C2': '88.70%', 'C1': '11.30%'}

{'F8': 'coarseaggregate', 'F6': 'age_days', 'F1': 'superplasticizer', 'F7': 'cement', 'F3': 'water', 'F4': 'fineaggregate', 'F2': 'slag', 'F5': 'flyash'}

{'F6': 'F8', 'F8': 'F6', 'F5': 'F1', 'F1': 'F7', 'F4': 'F3', 'F7': 'F4', 'F2': 'F2', 'F3': 'F5'}

{'C1': 'C2', 'C2': 'C1'}

Weak

{'C2': 'Weak', 'C1': 'Strong'}

SVC

C2

Broadband Sevice Signup

With a higher level of certainty, the algorithm labels the given data or case as C2 because the predicted probability of class C2 is 99.93% while that of class C1 is only 0.07%. C1 is therefore less likely than C2 and the classification assertion or decision here is chiefly attributed to the impact of input features such as F16, F6, F39, F35, and F32. Among these relevant features, only F32 has a negative contribution, mildly dragging the verdict in favour of C1, whereas conversely, F16, F6, F39, and F35 have strong positive contributions in support of assigning C2 to the given data. Other features with moderate influence on the algorithm's verdict here include F3, F33, F10, F21, F22, F19, F24, and F4. However, some of the input features are shown to have negligible contribution to the abovementioned classification output and in fact, these include F1, F27, F30, and F9. In summary, the most vital features with respect to this classification instance are F16, F39, and F6 with positive contributions strongly increasing the algorithm's response towards label C2 hence the 99.93% predicted probability.

{'C2': '99.93%', 'C1': '0.07%'}

{'F16': 'X38', 'F6': 'X32', 'F39': 'X31', 'F35': 'X25', 'F32': 'X8', 'F3': 'X35', 'F33': 'X1', 'F24': 'X3', 'F21': 'X28', 'F4': 'X19', 'F22': 'X9', 'F10': 'X11', 'F19': 'X10', 'F37': 'X21', 'F12': 'X17', 'F38': 'X4', 'F5': 'X36', 'F8': 'X2', 'F11': 'X6', 'F29': 'X34', 'F27': 'X37', 'F1': 'X40', 'F30': 'X42', 'F9': 'X41', 'F42': 'X5', 'F28': 'X33', 'F34': 'X39', 'F7': 'X24', 'F40': 'X30', 'F20': 'X27', 'F2': 'X26', 'F41': 'X23', 'F26': 'X22', 'F18': 'X20', 'F13': 'X18', 'F36': 'X16', 'F14': 'X15', 'F23': 'X14', 'F25': 'X13', 'F17': 'X12', 'F31': 'X7', 'F15': 'X29'}

{'F35': 'F16', 'F29': 'F6', 'F28': 'F39', 'F23': 'F35', 'F6': 'F32', 'F32': 'F3', 'F40': 'F33', 'F2': 'F24', 'F26': 'F21', 'F17': 'F4', 'F7': 'F22', 'F9': 'F10', 'F8': 'F19', 'F19': 'F37', 'F15': 'F12', 'F3': 'F38', 'F33': 'F5', 'F1': 'F8', 'F4': 'F11', 'F31': 'F29', 'F34': 'F27', 'F37': 'F1', 'F38': 'F30', 'F39': 'F9', 'F41': 'F42', 'F30': 'F28', 'F36': 'F34', 'F22': 'F7', 'F27': 'F40', 'F25': 'F20', 'F24': 'F2', 'F21': 'F41', 'F20': 'F26', 'F18': 'F18', 'F16': 'F13', 'F14': 'F36', 'F13': 'F14', 'F12': 'F23', 'F11': 'F25', 'F10': 'F17', 'F5': 'F31', 'F42': 'F15'}

{'C1': 'C2', 'C2': 'C1'}

No

{'C2': 'No', 'C1': 'Yes'}

KNeighborsClassifier

C2

E-Commerce Shipping

There is uncertainty about the correct label for the given example since both labels, C2 and C1 are shown to have a 50.0% chance of being correct. The prediction decision above is mainly attributed to the influence of the input features F1, F8, and F6, while F4, F2, and F5 are deemed less important to the decision above. Looking at the direction of influence of each input feature, only F8, F8, F4, and F5 are shown to have a positive contribution, increasing the model's response towards assigning C2. All the remaining six features have a negative contribution towards the decision here, supporting the assignment of the other class.

{'C2': '50.00%', 'C1': '50.00%'}

{'F1': 'Discount_offered', 'F8': 'Weight_in_gms', 'F6': 'Prior_purchases', 'F3': 'Customer_care_calls', 'F10': 'Product_importance', 'F9': 'Mode_of_Shipment', 'F7': 'Warehouse_block', 'F4': 'Cost_of_the_Product', 'F2': 'Customer_rating', 'F5': 'Gender'}

{'F2': 'F1', 'F3': 'F8', 'F8': 'F6', 'F6': 'F3', 'F9': 'F10', 'F5': 'F9', 'F4': 'F7', 'F1': 'F4', 'F7': 'F2', 'F10': 'F5'}

{'C1': 'C2', 'C2': 'C1'}

On-time

{'C2': 'On-time', 'C1': 'Late'}

DecisionTreeClassifier

C2

Credit Risk Classification

The classification model assigned the label C2 to the given example and given that the confidence level is 100.0%, we can be certain that the chances of C1 being the true label are negligible. The most relevant features controlling the prediction decision above are F4, F5, and F3. F9, F6, and F1 are among the least relevant features. Most of the properties have values that sway the decision towards the other C1 class. The only features that increase the odds that C2 is the correct label are F4, F10, and F6. It is strange that the model has 100.0% confidence in its prediction for the selected sample, given that only a small number of the input features contribute positively to reaching the C2 estimate.

{'C1': '0.00%', 'C2': '100.00%'}

{'F4': 'fea_4', 'F5': 'fea_8', 'F3': 'fea_5', 'F11': 'fea_2', 'F7': 'fea_1', 'F2': 'fea_9', 'F8': 'fea_11', 'F10': 'fea_6', 'F9': 'fea_10', 'F6': 'fea_7', 'F1': 'fea_3'}

{'F4': 'F4', 'F8': 'F5', 'F5': 'F3', 'F2': 'F11', 'F1': 'F7', 'F9': 'F2', 'F11': 'F8', 'F6': 'F10', 'F10': 'F9', 'F7': 'F6', 'F3': 'F1'}

{'C1': 'C1', 'C2': 'C2'}

High

{'C1': 'Low', 'C2': 'High'}

LogisticRegression

C1

Hotel Satisfaction

The model prediction for the test case is C1 and the confidence level of this is almost 100%. From examining the contributions of variables or attributes, the values of F6 and F13 push the prediction verdict in favor of the other label. On the contrary, F10, F14, F4, and F8 have values with a positive influence that biases the classification decision towards label C1. While attributes F5 and F12 contradict the prediction made, F15 and F2 have values that support the model's prediction for the given case.

{'C1': '91.36%', 'C2': '8.64%'}

{'F6': 'Type of Travel', 'F13': 'Type Of Booking', 'F10': 'Hotel wifi service', 'F14': 'Common Room entertainment', 'F4': 'Stay comfort', 'F8': 'Other service', 'F5': 'Checkin\\/Checkout service', 'F15': 'Hotel location', 'F12': 'Food and drink', 'F2': 'Cleanliness', 'F7': 'Age', 'F1': 'Departure\\/Arrival convenience', 'F11': 'purpose_of_travel', 'F3': 'Ease of Online booking', 'F9': 'Gender'}

{'F3': 'F6', 'F4': 'F13', 'F6': 'F10', 'F12': 'F14', 'F11': 'F4', 'F14': 'F8', 'F13': 'F5', 'F9': 'F15', 'F10': 'F12', 'F15': 'F2', 'F5': 'F7', 'F7': 'F1', 'F2': 'F11', 'F8': 'F3', 'F1': 'F9'}

{'C1': 'C1', 'C2': 'C2'}

dissatisfied

{'C1': 'dissatisfied', 'C2': 'satisfied'}

SGDClassifier

C1

Company Bankruptcy Prediction

The label predicted by the classifier is C1 at a 71.80% confidence level. On the other hand, there is a 28.20% chance that C2 could be the label. The prediction can be mainly attributed to contributions from F38, F59, F47, and F27. Considerable positive contributions to the prediction here are from F38, F27, F52, and F59 since their values support the prediction of C1. Shifting the prediction towards C2 are the negative features F47, F16, F75, F13, and F32. There were some features with minuscule influence on prediction decision made for the case under consideration; these include F17, F54, and F6. In simple terms, the classifer deems the values of these features less important when assigning the label here.

{'C1': '71.80%', 'C2': '28.20%'}

{'F38': ' Liability to Equity', 'F59': ' Net worth\\/Assets', 'F47': ' Debt ratio %', 'F27': ' Equity to Liability', 'F16': ' Realized Sales Gross Margin', 'F52': ' Net Value Per Share (A)', 'F75': ' Net Income to Total Assets', 'F13': ' Current Liability to Equity', 'F68': ' Current Liability to Assets', 'F32': ' Current Liabilities\\/Equity', 'F57': ' Quick Ratio', 'F25': ' Net Value Per Share (C)', 'F10': ' Working Capital to Total Assets', 'F53': ' Operating Funds to Liability', 'F61': ' Total expense\\/Assets', 'F21': ' Current Liability to Current Assets', 'F76': ' Quick Assets\\/Current Liability', 'F64': ' Continuous Net Profit Growth Rate', 'F31': ' Total debt\\/Total net worth', 'F26': ' After-tax Net Profit Growth Rate', 'F93': ' Total income\\/Total expense', 'F65': ' Operating Profit Rate', 'F6': " Net Income to Stockholder's Equity", 'F7': ' Cash Flow to Equity', 'F56': ' Interest Coverage Ratio (Interest expense to EBIT)', 'F72': ' Current Asset Turnover Rate', 'F63': ' Cash Flow Per Share', 'F24': ' Operating Gross Margin', 'F81': ' Operating Profit Per Share (Yuan ¥)', 'F45': ' Contingent liabilities\\/Net worth', 'F17': ' Net Worth Turnover Rate (times)', 'F5': ' No-credit Interval', 'F88': ' Net profit before tax\\/Paid-in capital', 'F71': ' Working Capital\\/Equity', 'F78': ' Per Share Net profit before tax (Yuan ¥)', 'F50': ' Current Liability to Liability', 'F12': ' Operating profit\\/Paid-in capital', 'F54': ' Regular Net Profit Growth Rate', 'F66': ' Current Ratio', 'F19': ' Tax rate (A)', 'F39': ' After-tax net Interest Rate', 'F33': ' Total Asset Turnover', 'F44': ' Long-term Liability to Current Assets', 'F3': ' CFO to Assets', 'F85': ' Cash Reinvestment %', 'F70': ' Fixed Assets to Assets', 'F9': ' Working capitcal Turnover Rate', 'F41': ' Current Liabilities\\/Liability', 'F15': ' Inventory and accounts receivable\\/Net value', 'F89': ' Long-term fund suitability ratio (A)', 'F22': ' Interest-bearing debt interest rate', 'F46': ' Cash Flow to Liability', 'F11': ' Interest Expense Ratio', 'F58': ' Equity to Long-term Liability', 'F8': ' Fixed Assets Turnover Frequency', 'F91': ' Inventory\\/Current Liability', 'F84': ' Allocation rate per person', 'F82': ' Operating Expense Rate', 'F79': ' Inventory Turnover Rate (times)', 'F77': ' Operating profit per person', 'F87': ' Net Value Growth Rate', 'F69': ' ROA(B) before interest and depreciation after tax', 'F42': ' Cash Flow to Total Assets', 'F18': ' Continuous interest rate (after tax)', 'F86': ' Inventory\\/Working Capital', 'F60': ' Retained Earnings to Total Assets', 'F40': ' Total assets to GNP price', 'F1': ' Persistent EPS in the Last Four Seasons', 'F92': ' Revenue per person', 'F74': ' Non-industry income and expenditure\\/revenue', 'F29': ' Borrowing dependency', 'F37': ' Total Asset Growth Rate', 'F35': ' Cash Flow to Sales', 'F73': ' Cash\\/Total Assets', 'F55': ' Net Value Per Share (B)', 'F2': ' Pre-tax net Interest Rate', 'F20': ' Accounts Receivable Turnover', 'F62': ' Quick Assets\\/Total Assets', 'F51': ' Operating Profit Growth Rate', 'F83': ' Average Collection Days', 'F90': ' Current Assets\\/Total Assets', 'F14': ' Realized Sales Gross Profit Growth Rate', 'F43': ' Cash flow rate', 'F49': ' Total Asset Return Growth Rate Ratio', 'F67': ' Degree of Financial Leverage (DFL)', 'F36': ' Cash Turnover Rate', 'F28': ' Quick Asset Turnover Rate', 'F80': ' Cash\\/Current Liability', 'F48': ' Revenue Per Share (Yuan ¥)', 'F23': ' Research and development expense rate', 'F4': ' ROA(C) before interest and depreciation before interest', 'F30': ' ROA(A) before interest and % after tax', 'F34': ' Gross Profit to Sales'}

{'F66': 'F38', 'F84': 'F59', 'F47': 'F47', 'F91': 'F27', 'F83': 'F16', 'F42': 'F52', 'F16': 'F75', 'F92': 'F13', 'F46': 'F68', 'F39': 'F32', 'F6': 'F57', 'F88': 'F25', 'F67': 'F10', 'F87': 'F53', 'F44': 'F61', 'F86': 'F21', 'F71': 'F76', 'F54': 'F64', 'F7': 'F31', 'F80': 'F26', 'F57': 'F93', 'F58': 'F65', 'F59': 'F6', 'F53': 'F7', 'F60': 'F56', 'F61': 'F72', 'F65': 'F63', 'F62': 'F24', 'F63': 'F81', 'F64': 'F45', 'F55': 'F17', 'F56': 'F5', 'F72': 'F88', 'F68': 'F71', 'F78': 'F78', 'F90': 'F50', 'F89': 'F12', 'F85': 'F54', 'F82': 'F66', 'F81': 'F19', 'F79': 'F39', 'F77': 'F33', 'F69': 'F44', 'F76': 'F3', 'F75': 'F85', 'F74': 'F70', 'F73': 'F9', 'F51': 'F41', 'F70': 'F15', 'F52': 'F89', 'F1': 'F22', 'F50': 'F46', 'F14': 'F11', 'F23': 'F58', 'F22': 'F8', 'F21': 'F91', 'F20': 'F84', 'F19': 'F82', 'F18': 'F79', 'F17': 'F77', 'F15': 'F87', 'F13': 'F69', 'F49': 'F42', 'F12': 'F18', 'F11': 'F86', 'F10': 'F60', 'F9': 'F40', 'F8': 'F1', 'F5': 'F92', 'F4': 'F74', 'F3': 'F29', 'F24': 'F37', 'F25': 'F35', 'F26': 'F73', 'F27': 'F55', 'F48': 'F2', 'F2': 'F20', 'F45': 'F62', 'F43': 'F51', 'F41': 'F83', 'F40': 'F90', 'F38': 'F14', 'F37': 'F43', 'F36': 'F49', 'F35': 'F67', 'F34': 'F36', 'F33': 'F28', 'F32': 'F80', 'F31': 'F48', 'F30': 'F23', 'F29': 'F4', 'F28': 'F30', 'F93': 'F34'}

{'C1': 'C1', 'C2': 'C2'}

No

{'C1': 'No', 'C2': 'Yes'}

BernoulliNB

C2

Employee Promotion Prediction

The model, making a classification decision based on the input variables, predicts the class C2 label for this case with a predicted likelihood equal to 54.21%. It also shows a 45.79% probability that C1 is the correct label. The classification decision made above is primarily influenced by the variables F2, F4, F1, F8, and F10. The three most influential variables, F2, F4, and F8, have a negative impact since their values are shifting the labelling decision in the direction of C1 instead of C2. Positive variables are F10, F1, F11, F7, and F6, supporting the model's class assignment decision for this situation and one can conclude that it is the influence of the positives that motivates the decision towards C2.

{'C2': '54.21%', 'C1': '45.79%'}

{'F2': 'KPIs_met >80%', 'F4': 'previous_year_rating', 'F8': 'avg_training_score', 'F10': 'department', 'F1': 'education', 'F11': 'recruitment_channel', 'F9': 'no_of_trainings', 'F3': 'length_of_service', 'F7': 'region', 'F5': 'age', 'F6': 'gender'}

{'F10': 'F2', 'F8': 'F4', 'F11': 'F8', 'F1': 'F10', 'F3': 'F1', 'F5': 'F11', 'F6': 'F9', 'F9': 'F3', 'F2': 'F7', 'F7': 'F5', 'F4': 'F6'}

{'C1': 'C2', 'C2': 'C1'}

Ignore

{'C2': 'Ignore', 'C1': 'Promote'}

MLPClassifier

C2

Vehicle Insurance Claims

Based on the values of the input features, the classifier believes that the most probable label for the given data is C2, due to the fact that there is only a 19.30% chance that it could be C1 instead. The most influential features resulting in the decision or judgement above are F9, F24, F14, F6, F3, F30, and F22, though features such as F12, F31, F19, and F5 are indicated to have negligible contributions to the classification. Actually, the high certainty of the chosen label can be attributed to the very strong positive influence of F9 and the moderate positive influence of F24, F14, F11, and F6. Conversely, the negative features F30, F22, F3, and F18 reduce the likelihood of C2 since their values support labelling the case as C1.

{'C1': '19.30%', 'C2': '80.70%'}

{'F9': 'incident_severity', 'F24': 'insured_relationship', 'F6': 'authorities_contacted', 'F14': 'vehicle_claim', 'F30': 'umbrella_limit', 'F22': 'insured_hobbies', 'F3': 'incident_type', 'F11': 'policy_deductable', 'F18': 'auto_make', 'F23': 'number_of_vehicles_involved', 'F33': 'insured_occupation', 'F16': 'property_damage', 'F29': 'incident_state', 'F20': 'auto_year', 'F15': 'capital-loss', 'F28': 'policy_csl', 'F25': 'collision_type', 'F13': 'capital-gains', 'F21': 'property_claim', 'F8': 'incident_hour_of_the_day', 'F12': 'police_report_available', 'F31': 'policy_annual_premium', 'F19': 'incident_city', 'F5': 'insured_zip', 'F7': 'bodily_injuries', 'F4': 'injury_claim', 'F2': 'witnesses', 'F1': 'total_claim_amount', 'F27': 'insured_education_level', 'F32': 'insured_sex', 'F26': 'policy_state', 'F17': 'age', 'F10': 'months_as_customer'}

{'F27': 'F9', 'F24': 'F24', 'F28': 'F6', 'F16': 'F14', 'F5': 'F30', 'F23': 'F22', 'F25': 'F3', 'F3': 'F11', 'F33': 'F18', 'F10': 'F23', 'F22': 'F33', 'F31': 'F16', 'F29': 'F29', 'F17': 'F20', 'F8': 'F15', 'F19': 'F28', 'F26': 'F25', 'F7': 'F13', 'F15': 'F21', 'F9': 'F8', 'F32': 'F12', 'F4': 'F31', 'F30': 'F19', 'F6': 'F5', 'F11': 'F7', 'F14': 'F4', 'F12': 'F2', 'F13': 'F1', 'F21': 'F27', 'F20': 'F32', 'F18': 'F26', 'F2': 'F17', 'F1': 'F10'}

{'C1': 'C1', 'C2': 'C2'}

Fraud

{'C1': 'Not Fraud', 'C2': 'Fraud'}

SGDClassifier

C2

House Price Classification

According to the classification algorithm with a very high confidence level, the correct label for the given data instance is C2. This prediction decision is heavily influenced by features such as F1, F6, F2, F10, F13, and F9. Among these top features, the only features with a negative contribution towards the assigned label are F9 and F13. With respect to the given instance, their negative contributions decrease the algorithm's response in favour of the least probable class. F5, F3, F11, F12, and F8 positively support the assignment of label C2. Conversely, F7 and F3 have a similar direction of influence as F13 and F9.

{'C1': '0.00%', 'C2': '100.00%'}

{'F1': 'AGE', 'F6': 'RAD', 'F2': 'LSTAT', 'F10': 'RM', 'F9': 'DIS', 'F13': 'CHAS', 'F5': 'ZN', 'F3': 'CRIM', 'F7': 'TAX', 'F11': 'B', 'F12': 'PTRATIO', 'F8': 'INDUS', 'F4': 'NOX'}

{'F7': 'F1', 'F9': 'F6', 'F13': 'F2', 'F6': 'F10', 'F8': 'F9', 'F4': 'F13', 'F2': 'F5', 'F1': 'F3', 'F10': 'F7', 'F12': 'F11', 'F11': 'F12', 'F3': 'F8', 'F5': 'F4'}

{'C1': 'C1', 'C2': 'C2'}

High

{'C1': 'Low', 'C2': 'High'}

RandomForestClassifier

C2

Bike Sharing Demand

For the given data instance, the most probable class according to the classifier is C2 since the probability of C1 being the correct label is only about 10.0%. The most influential features resulting in the prediction decision above are F8, F1, and F7 which are shown to negatively contribute to the decision above since they strongly push the classifier towards assigning a different label. F5, F9, and F4 are shown to be the only features to positively contribute to the classification made here. Aside from the positive features, all the others negatively reduce the odds of the given data instance having C2 as its label.

{'C2': '90.00%', 'C1': '10.00%'}

{'F8': 'Functioning Day', 'F1': 'Rainfall(mm)', 'F7': 'Temperature', 'F5': 'Solar Radiation (MJ\\/m2)', 'F9': 'Seasons', 'F12': 'Wind speed (m\\/s)', 'F4': 'Holiday', 'F3': 'Visibility (10m)', 'F6': 'Dew point temperature', 'F11': 'Hour', 'F2': 'Snowfall (cm)', 'F10': 'Humidity(%)'}

{'F12': 'F8', 'F8': 'F1', 'F2': 'F7', 'F7': 'F5', 'F10': 'F9', 'F4': 'F12', 'F11': 'F4', 'F5': 'F3', 'F6': 'F6', 'F1': 'F11', 'F9': 'F2', 'F3': 'F10'}

{'C1': 'C2', 'C2': 'C1'}

Less than 500

{'C2': 'Less than 500', 'C1': 'More than 500'}

RandomForestClassifier

C1

Advertisement Prediction

Judging based on the information about the given case, the model outputs C1 with a prediction probability of 74.72%, however, it is vital to keep in mind that there is also a 25.28% probability that C2 could be the true label. The attribution analysis shows that all the input variables have varying degrees of influence on the model as it arrives at the abovementioned decision and the influence of the features can be ranked from the most relevant to the least relevant as follows: F5, F1, F3, F7, F6, F4, and F2. Across the input features, only F1 and F7 have negative attributions, reducing the likelihood of the predicted label which explain the 25.28% predicted likelihood of the C2 label. Therefore, F5, F3, F6, F4, and F2 are the positive input features pushing the decision higher towards C1 and away from C2.

{'C1': '74.72%', 'C2': '25.28%'}

{'F5': 'Daily Time Spent on Site', 'F1': 'Daily Internet Usage', 'F3': 'Age', 'F7': 'ad_day', 'F6': 'Area Income', 'F4': 'Gender', 'F2': 'ad_month'}

{'F1': 'F5', 'F4': 'F1', 'F2': 'F3', 'F7': 'F7', 'F3': 'F6', 'F5': 'F4', 'F6': 'F2'}

{'C1': 'C1', 'C2': 'C2'}

Skip

{'C1': 'Skip', 'C2': 'Watch'}

DNN

C2

Credit Card Fraud Classification

The data is labelled C2 by the model as it has a somewhat greater prediction chance than C1. F6, F5, F10, F13, and F28 are the input variables that have the most impact on the above classification choice, whereas F14, F19, F30, F9, and F12 have the least influence. F6, F5, F28, and F13 are basically supporting the choice of the label in this scenario while on the contrary, F10, F18, and F1 are the primary negative factors. It's not unexpected that the model isn't 100 percent sure of the assigned label considering the degree of influence as well as the direction of influence of the variables.

{'C1': '48.58%', 'C2': '51.42%'}

{'F5': 'Z18', 'F6': 'Z14', 'F10': 'Time', 'F13': 'Z1', 'F28': 'Z19', 'F3': 'Z10', 'F25': 'Z4', 'F22': 'Z3', 'F23': 'Z12', 'F7': 'Z16', 'F18': 'Z7', 'F1': 'Z11', 'F26': 'Z9', 'F8': 'Z6', 'F2': 'Z23', 'F4': 'Z5', 'F15': 'Z17', 'F29': 'Z21', 'F20': 'Z24', 'F21': 'Z8', 'F24': 'Amount', 'F27': 'Z20', 'F16': 'Z27', 'F11': 'Z25', 'F17': 'Z13', 'F14': 'Z2', 'F19': 'Z22', 'F30': 'Z28', 'F9': 'Z26', 'F12': 'Z15'}

{'F19': 'F5', 'F15': 'F6', 'F1': 'F10', 'F2': 'F13', 'F20': 'F28', 'F11': 'F3', 'F5': 'F25', 'F4': 'F22', 'F13': 'F23', 'F17': 'F7', 'F8': 'F18', 'F12': 'F1', 'F10': 'F26', 'F7': 'F8', 'F24': 'F2', 'F6': 'F4', 'F18': 'F15', 'F22': 'F29', 'F25': 'F20', 'F9': 'F21', 'F30': 'F24', 'F21': 'F27', 'F28': 'F16', 'F26': 'F11', 'F14': 'F17', 'F3': 'F14', 'F23': 'F19', 'F29': 'F30', 'F27': 'F9', 'F16': 'F12'}

{'C1': 'C1', 'C2': 'C2'}

Fraud

{'C1': 'Not Fraud', 'C2': 'Fraud'}

GradientBoostingClassifier

C1

Printer Sales

Although the case under consideration has variables with a significant negative impact, it also has many measurable variables that are positive, so there is a good chance that C1 is correct since it has a 91.95% certainty. F15, F2, and F6 are the most important input variables, thanks to which the model successfully assigns the selected label, C1. F22 and F19 have almost identical positive impacts, while F4 has negative effects, shifting the output decision in favour of a different label. However, the cjoint positive contributions of F22, F15, F6, and F19 was higher than that of F2, F9, F18, and F4, increasing the likelihood of the C1 class. Unfortunately, the values of the variables F17, F26, F5, and F1 are likely ignored since their attributions are much closer to zero.

{'C2': '8.05%', 'C1': '91.95%'}

{'F15': 'X24', 'F2': 'X8', 'F6': 'X1', 'F9': 'X21', 'F18': 'X4', 'F19': 'X6', 'F22': 'X3', 'F4': 'X22', 'F8': 'X7', 'F10': 'X15', 'F11': 'X20', 'F7': 'X11', 'F23': 'X10', 'F25': 'X19', 'F3': 'X5', 'F12': 'X16', 'F24': 'X23', 'F14': 'X9', 'F21': 'X17', 'F20': 'X18', 'F13': 'X25', 'F16': 'X14', 'F1': 'X2', 'F26': 'X13', 'F17': 'X12', 'F5': 'X26'}

{'F24': 'F15', 'F8': 'F2', 'F1': 'F6', 'F21': 'F9', 'F4': 'F18', 'F6': 'F19', 'F3': 'F22', 'F22': 'F4', 'F7': 'F8', 'F15': 'F10', 'F20': 'F11', 'F11': 'F7', 'F10': 'F23', 'F19': 'F25', 'F5': 'F3', 'F16': 'F12', 'F23': 'F24', 'F9': 'F14', 'F17': 'F21', 'F18': 'F20', 'F25': 'F13', 'F14': 'F16', 'F2': 'F1', 'F13': 'F26', 'F12': 'F17', 'F26': 'F5'}

{'C1': 'C2', 'C2': 'C1'}

More

{'C2': 'Less', 'C1': 'More'}

LogisticRegression

C3

Air Quality Prediction

The classification output observations that follow are based on the information supplied about this specific case. The class label in this case is forecasted to be C3 out of the four possible labels, with a probability of around 83.08 percent. With a probability of 16.87 percent, C2 is the next most likely label. The third possible label, C1, has a 0.05 percent chance of being correct. The algorithm, on the other hand, confirms that C4 is unlikely to be the correct label. According to the attribution analysis, F4, F3, F1, F5, F6, F2 is the ranking of the input features based on how powerful their effect on the algorithm is. Furthermore, among the input variables, F4 and F5 exhibit negative attributions, causing the decision to be shifted away from label C3. Finally, F3, F1, F6, and F2 are the positive variables that sway the judgement in favour of C3.

{'C3': '83.08%', 'C2': '16.87%', 'C1': '0.00%', 'C4': '0.05%'}

{'F4': 'MQ5', 'F3': 'MQ3', 'F1': 'MQ1', 'F5': 'MQ4', 'F6': 'MQ6', 'F2': 'MQ2'}

{'F5': 'F4', 'F3': 'F3', 'F1': 'F1', 'F4': 'F5', 'F6': 'F6', 'F2': 'F2'}

{'C1': 'C3', 'C2': 'C2', 'C3': 'C1', 'C4': 'C4'}

Preparing meals

{'C3': 'Preparing meals', 'C2': 'Presence of smoke', 'C1': 'Cleaning', 'C4': 'Other'}

DecisionTreeClassifier

C2

Insurance Churn

The model predicted class C2 with a very high confidence level of 93.27% and looking at the predicted probabilities across the label, there is only a 6.73% chance that C1 is the true label. There are two features that have a very strong positive effect on the prediction of class C2 and these are F12 and F4. The following features have moderate impact and are listed in descending order of impact: F16 and F9 have a negative impact, while F3 and F8 have a positive impact on the prediction of C2. In addition, both F5 and F10 had a negative effect on the model, further decreasing the odds of C2 being the true label for the given case. Finally, in terms of model decisions for this case, the features with the least contributions are F14, F1, F11, and F6.

{'C1': '6.73%', 'C2': '93.27%'}

{'F12': 'feature15', 'F4': 'feature14', 'F16': 'feature10', 'F9': 'feature11', 'F3': 'feature5', 'F8': 'feature13', 'F5': 'feature4', 'F10': 'feature3', 'F2': 'feature12', 'F7': 'feature1', 'F13': 'feature7', 'F15': 'feature2', 'F14': 'feature6', 'F1': 'feature0', 'F11': 'feature9', 'F6': 'feature8'}

{'F9': 'F12', 'F8': 'F4', 'F4': 'F16', 'F5': 'F9', 'F15': 'F3', 'F7': 'F8', 'F14': 'F5', 'F13': 'F10', 'F6': 'F2', 'F11': 'F7', 'F1': 'F13', 'F12': 'F15', 'F16': 'F14', 'F10': 'F1', 'F3': 'F11', 'F2': 'F6'}

{'C1': 'C1', 'C2': 'C2'}

Leave

{'C1': 'Stay', 'C2': 'Leave'}

RandomForestClassifier

C3

Flight Price-Range Classification

Between the three possible classes, there is a 100% certainty that the correct label for this case is C3. The features with a very high impact on the prediction made here are F7, F6, and F2, which are also shown to have a very strong positive contribution to the C3 prediction. Other features that shift the prediction in favour of C3 are F10, F11, F3, F12, and F8. On the other hand, F5, F4, and F9 negatively swing the model towards predicting a different label. Compared to F7, F6, and F2, all the negative features have a low to moderate influence on the prediction made here. Finally, F1 has the lowest positive contribution that also further increases the likelihood of the output label, C3.

{'C3': '100.00%', 'C1': '0.00%', 'C2': '0.00%'}

{'F7': 'Duration_hours', 'F2': 'Airline', 'F6': 'Total_Stops', 'F10': 'Journey_day', 'F5': 'Source', 'F11': 'Destination', 'F3': 'Journey_month', 'F12': 'Dep_minute', 'F8': 'Arrival_minute', 'F4': 'Arrival_hour', 'F9': 'Duration_mins', 'F1': 'Dep_hour'}

{'F7': 'F7', 'F9': 'F2', 'F12': 'F6', 'F1': 'F10', 'F10': 'F5', 'F11': 'F11', 'F2': 'F3', 'F4': 'F12', 'F6': 'F8', 'F5': 'F4', 'F8': 'F9', 'F3': 'F1'}

{'C1': 'C3', 'C2': 'C1', 'C3': 'C2'}

Low

{'C3': 'Low', 'C1': 'Moderate', 'C2': 'High'}

SVC

C2

Advertisement Prediction

For the given case or instance, the model assigns the label C2, with the prediction confidence equal to 56.56%. The variables F4, F1, F5, and F6 all contribute a lot to the classification decision above. While F4 and F5 are impacting positively, F1 and F6 are decreasing the likelihood of the assigned label. For the remaining features, both F3 and F7 shift the classification towards C2, whereas F2 has a marginal influence on the model, shifting the final verdict away in favour of the alternative label.

{'C2': '56.56%', 'C1': '43.44%'}

{'F4': 'Daily Time Spent on Site', 'F1': 'Daily Internet Usage', 'F5': 'Age', 'F6': 'ad_day', 'F3': 'ad_month', 'F7': 'Area Income', 'F2': 'Gender'}

{'F1': 'F4', 'F4': 'F1', 'F2': 'F5', 'F7': 'F6', 'F6': 'F3', 'F3': 'F7', 'F5': 'F2'}

{'C1': 'C2', 'C2': 'C1'}

Skip

{'C2': 'Skip', 'C1': 'Watch'}

RandomForestClassifier

C1

Paris House Classification

The prediction made for this case by the model is that C1 is most likely the true label, with a confidence level of 72.03% higher than the 27.97% of the C2 label. According to the input features attribution analysis conducted, the features with the most influence on the decision are F1, F8, F15, and F10, all of which increase the probability that C1 is indeed the true label. The top negatively contributing features, increasing the probability that perhaps the true label could be C2, on the other hand, are F11, F17, F5, and F4. Conversely, F9, F6, F2, and F14 also have positive contributions, further pushing the decision towards labelling the case as C1. Overall, the fairly high confidence in the classification decision here can be attributed to the fact that positive features have a much higher influence on the decision than their negative counterparts.

{'C2': '27.97%', 'C1': '72.03%'}

{'F8': 'isNewBuilt', 'F1': 'hasYard', 'F15': 'hasPool', 'F10': 'hasStormProtector', 'F11': 'made', 'F17': 'hasGuestRoom', 'F5': 'floors', 'F9': 'squareMeters', 'F6': 'numPrevOwners', 'F4': 'cityCode', 'F7': 'price', 'F12': 'numberOfRooms', 'F3': 'basement', 'F2': 'attic', 'F13': 'cityPartRange', 'F14': 'hasStorageRoom', 'F16': 'garage'}

{'F3': 'F8', 'F1': 'F1', 'F2': 'F15', 'F4': 'F10', 'F12': 'F11', 'F16': 'F17', 'F8': 'F5', 'F6': 'F9', 'F11': 'F6', 'F9': 'F4', 'F17': 'F7', 'F7': 'F12', 'F13': 'F3', 'F14': 'F2', 'F10': 'F13', 'F5': 'F14', 'F15': 'F16'}

{'C1': 'C2', 'C2': 'C1'}

Luxury

{'C2': 'Basic', 'C1': 'Luxury'}

LogisticRegression

C1

Annual Income Earnings

Tasked with labelling cases, the classification model labels the case under consideration as C1 since the probability of C2 is only 20.22%. The predicted probability of the less probable class, C2, reflects the fact that the model is a bit doubtful about the output label. Responsible for this doubt are the negative features F10, F14, F9, and F7 since they support labelling the given case as C2 over C1. On the contrary, F5, F3, F12, F8, F6, F4, F13, and F1 are among the positively contributing features, responsible for the moderately high confidence in the classification output decision.

{'C2': '20.22%', 'C1': '79.78%'}

{'F10': 'Capital Gain', 'F5': 'Marital Status', 'F3': 'Education', 'F14': 'Capital Loss', 'F12': 'Hours per week', 'F8': 'Sex', 'F6': 'Country', 'F1': 'Education-Num', 'F4': 'Occupation', 'F13': 'Race', 'F11': 'Age', 'F9': 'Workclass', 'F2': 'fnlwgt', 'F7': 'Relationship'}

{'F11': 'F10', 'F6': 'F5', 'F4': 'F3', 'F12': 'F14', 'F13': 'F12', 'F10': 'F8', 'F14': 'F6', 'F5': 'F1', 'F7': 'F4', 'F9': 'F13', 'F1': 'F11', 'F2': 'F9', 'F3': 'F2', 'F8': 'F7'}

{'C1': 'C2', 'C2': 'C1'}

Above 50K

{'C2': 'Under 50K', 'C1': 'Above 50K'}

SVM_poly

C1

Mobile Price-Range Classification

The classification algorithm determines that neither C3 nor C4 nor C2 is a suitable label for the present context. C1 is quite guaranteed to be the correct label. The aforementioned conclusion has a higher degree of confidence due to the positive contributions of F18, F20, and F11. Aside from the above mentioned positive variables, F14, F15, F8, and F10 are also positive. However, their influences are moderate compared to F18, F20, and F11 . The remaining positive variables, F6, F7, F9, and F3, are among the algorithm's least influential input variables. Other attributes, such as F13, F16, F19, and F1, merely serve to reduce the likelihood of C1 being the proper label in the current context. Given the algorithm's high confidence in this classification, one may conclude that the negative variables had minimal impact on the algorithm's label selection here.

{'C3': '0.00%', 'C4': '0.00%', 'C2': '0.00%', 'C1': '100.00%'}

{'F18': 'ram', 'F20': 'battery_power', 'F11': 'px_width', 'F13': 'int_memory', 'F16': 'sc_h', 'F19': 'wifi', 'F14': 'fc', 'F1': 'three_g', 'F15': 'mobile_wt', 'F4': 'clock_speed', 'F2': 'm_dep', 'F12': 'n_cores', 'F8': 'pc', 'F17': 'touch_screen', 'F10': 'blue', 'F6': 'talk_time', 'F5': 'sc_w', 'F7': 'px_height', 'F9': 'four_g', 'F3': 'dual_sim'}

{'F11': 'F18', 'F1': 'F20', 'F10': 'F11', 'F4': 'F13', 'F12': 'F16', 'F20': 'F19', 'F3': 'F14', 'F18': 'F1', 'F6': 'F15', 'F2': 'F4', 'F5': 'F2', 'F7': 'F12', 'F8': 'F8', 'F19': 'F17', 'F15': 'F10', 'F14': 'F6', 'F13': 'F5', 'F9': 'F7', 'F17': 'F9', 'F16': 'F3'}

{'C1': 'C3', 'C2': 'C4', 'C3': 'C2', 'C4': 'C1'}

r4

{'C3': 'r1', 'C4': 'r2', 'C2': 'r3', 'C1': 'r4'}

SVM_linear

C1

Employee Promotion Prediction

The model gave the output label as C1 with a very high probability of 99.69%, leaving only 0.31% chance that C2 could be the right one. According to the contributions or attributions analysis done to understand the properties of various traits, F10 is by far the most influential trait. F8 had a positive impact on model predictions, as did F3. This is in contrast to F6 and F4, which have a negative impact on the model, pushing the classification verdict towards C2. Several input features are shown to have a limited impact on the output label produced by the model and they are: F11, F2, F5, F7, and F1. Overall, only the features F9, F6, F4, F11, F2, and F7 showed negative attributions, reducing the likelihood of the C1 label being assigned by the model but their joint impact was not enough to predispose the model toward a different classification decision.

{'C2': '0.31%', 'C1': '99.69%'}

{'F10': 'avg_training_score', 'F9': 'department', 'F8': 'KPIs_met >80%', 'F6': 'recruitment_channel', 'F4': 'age', 'F3': 'no_of_trainings', 'F11': 'previous_year_rating', 'F2': 'education', 'F5': 'region', 'F7': 'length_of_service', 'F1': 'gender'}

{'F11': 'F10', 'F1': 'F9', 'F10': 'F8', 'F5': 'F6', 'F7': 'F4', 'F6': 'F3', 'F8': 'F11', 'F3': 'F2', 'F2': 'F5', 'F9': 'F7', 'F4': 'F1'}

{'C1': 'C2', 'C2': 'C1'}

Promote

{'C2': 'Ignore', 'C1': 'Promote'}

RandomForestClassifier

C2

Credit Risk Classification

Between the two classes, the given case is assigned the label C2 given that it has the highest predicted probability of about 93.0% since the probability of having C1 as the label is only 7.0%. Analysing the prediction made for the case under consideration, F6, F9, F4, and F10 are the features mainly pushing the prediction higher away from C2, while F5, F8, F3, and F1 improve the odds of the prediction being equal to C2. All things considered, the most relevant feature is F5, by contrast F2 and F11 are the ranked as the least relevant for the label assignment above.

{'C2': '93.00%', 'C1': '7.00%'}

{'F5': 'fea_4', 'F6': 'fea_10', 'F8': 'fea_8', 'F9': 'fea_7', 'F1': 'fea_2', 'F3': 'fea_3', 'F4': 'fea_5', 'F10': 'fea_1', 'F7': 'fea_9', 'F2': 'fea_6', 'F11': 'fea_11'}

{'F4': 'F5', 'F10': 'F6', 'F8': 'F8', 'F7': 'F9', 'F2': 'F1', 'F3': 'F3', 'F5': 'F4', 'F1': 'F10', 'F9': 'F7', 'F6': 'F2', 'F11': 'F11'}

{'C1': 'C2', 'C2': 'C1'}

Low

{'C2': 'Low', 'C1': 'High'}

SVM_linear

C1

Wine Quality Prediction

The classifier says that C1 has a 67.54 percent chance of being the correct label for the given example or case; consequently the label C2 has a 33.46 percent chance of being the chosen class. The variables F2, F11, F9, and F1 have the most impact on the prediction judgement here. On the other hand, F8, F3, and F5 are seen as less relevant variables when determining the proper class. The variables F1, F10, F3, and F5 lower the probability of the assigned label C1 since they are negative variables favouring the C2 prediction decision. However, the other features' collective or joint attribution is strong enough to favour C1. In summary, F11, F2, and F9 are the most positive variables.

{'C2': '32.46%', 'C1': '67.54%'}

{'F2': 'residual sugar', 'F11': 'volatile acidity', 'F9': 'alcohol', 'F1': 'fixed acidity', 'F6': 'chlorides', 'F10': 'sulphates', 'F7': 'citric acid', 'F4': 'free sulfur dioxide', 'F8': 'density', 'F3': 'total sulfur dioxide', 'F5': 'pH'}

{'F4': 'F2', 'F2': 'F11', 'F11': 'F9', 'F1': 'F1', 'F5': 'F6', 'F10': 'F10', 'F3': 'F7', 'F6': 'F4', 'F8': 'F8', 'F7': 'F3', 'F9': 'F5'}

{'C1': 'C2', 'C2': 'C1'}

high quality

{'C2': 'low_quality', 'C1': 'high quality'}

SVC

C2

Australian Credit Approval

The classification algorithm labels the presented data as C2 with the degree of confidence equal to 81.43 percent, although there is an 18.57 percent possibility that C1 is the correct label. The positive effects and contributions of input variables F8, F12, and F9 are mostly used to assign C2 to a specific scenario. Furthermore, the bulk of the remaining input variables contribute positively, making label C2 even more predictable. The only variables with negative contributions are F11, F3, F10, and F1, which move the choice to C1 rather than C2. Comparing the negative attributions to the positive attributions illustrates why the algorithm is certain that C2 is the correct label here.

{'C1': '18.57%', 'C2': '81.43%'}

{'F8': 'A8', 'F12': 'A9', 'F9': 'A14', 'F13': 'A12', 'F5': 'A7', 'F7': 'A4', 'F2': 'A5', 'F6': 'A11', 'F11': 'A1', 'F3': 'A13', 'F14': 'A10', 'F10': 'A2', 'F4': 'A6', 'F1': 'A3'}

{'F8': 'F8', 'F9': 'F12', 'F14': 'F9', 'F12': 'F13', 'F7': 'F5', 'F4': 'F7', 'F5': 'F2', 'F11': 'F6', 'F1': 'F11', 'F13': 'F3', 'F10': 'F14', 'F2': 'F10', 'F6': 'F4', 'F3': 'F1'}

{'C1': 'C1', 'C2': 'C2'}

Class 2

{'C1': 'Class 1', 'C2': 'Class 2'}

RandomForestClassifier

C1

E-Commerce Shipping

The probability that the label is C1 is 51.62% and the probability that C2 is the correct label is 48.38%. For this case or example, the uncertainty of the model is mainly due to the direction of influence of the variables F1, F7, and F9. Reducing the chance that C1 is the correct label are variables F1, F9, F3, and F6. While F1, F9, and F3 have a strong negative impact, F6 has the least negative contribution. Per the attribution analysis, increasing the prediction probability of C1 are the variables F7, F10, and F2 which are supported by F5, F8, and F4 all with moderate positive influences on the classification decision made by the model.

{'C1': '51.62%', 'C2': '48.38%'}

{'F1': 'Discount_offered', 'F7': 'Weight_in_gms', 'F9': 'Customer_care_calls', 'F3': 'Product_importance', 'F2': 'Mode_of_Shipment', 'F10': 'Warehouse_block', 'F4': 'Cost_of_the_Product', 'F8': 'Gender', 'F5': 'Customer_rating', 'F6': 'Prior_purchases'}

{'F2': 'F1', 'F3': 'F7', 'F6': 'F9', 'F9': 'F3', 'F5': 'F2', 'F4': 'F10', 'F1': 'F4', 'F10': 'F8', 'F7': 'F5', 'F8': 'F6'}

{'C1': 'C1', 'C2': 'C2'}

On-time

{'C1': 'On-time', 'C2': 'Late'}