Datasets:

james-burton

/

textual-explanations-702010

like 0

LogisticRegression

C2

Used Cars Price-Range Prediction

The prediction model predicts C2 for the case under consideration since the likelihood of C1 which is equal to 30.05%, is lower than that of C2 and this verdict came about mainly based on the values of the input features passed to the model. F10, F7, and F1 are identified as the most influential features with higher impact on the model's labelling decision here and among them F10 and F7 have negative contributions decreasing the model's response towards the assigned label. Furthermore, F1, F6, and F3 have a positive impact on the model and in effect pushes the decision higher towards C2, while F4, F9, and F2 have identical direction of impact as that of F7 and F10. Finally, F8 is the least relevant feature, therefore, its negative attribution has little effect on the model in this case and also the positive influence of F5 further supports the assigned label.

{'C1': '30.05%', 'C2': '69.95%'}

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

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

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

High

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

DecisionTreeClassifier

C2

Credit Risk Classification

The model assigned the label C2 to the given instance since its associated likelihood is far higher than C1. The most relevant features controlling the prediction decision above are F1, F3, and F4. The less relevant ones include F5, F9, and F7. The majority of the features have values, swinging the verdict towards the other class, C1. The only features increasing the likelihood or probability of C2 being the correct label are F1, F8, and F9. Given that only few features positively contribute to arriving at the C2 prediction, it is very strange that the model has 100.0% confidence in its prediction for the selected instance.

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

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

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

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

High

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

RandomForestClassifier

C1

Music Concert Attendance

There is an 80.0% chance that the true label for the given case is C1. Nine out of twenty features have a positive impact. Most features have a moderately low positive or negative impact, with the exception of F2, F7, and F16 and it appears as if F2 has an extremely negative impact, while F7 and F16 have the greater positive impacts. F5 has positive impacts, whereas the attributions of the features F1 and F9 are negatives. The least important features include F19, F8, F15, F4, F10, F3, and F20 with varying smaller effects.

{'C1': '80.00%', 'C2': '20.00%'}

{'F2': 'X11', 'F7': 'X1', 'F16': 'X6', 'F1': 'X10', 'F5': 'X14', 'F9': 'X16', 'F14': 'X13', 'F18': 'X12', 'F17': 'X3', 'F19': 'X2', 'F8': 'X15', 'F15': 'X4', 'F4': 'X7', 'F10': 'X17', 'F3': 'X8', 'F20': 'X5', 'F12': 'X18', 'F13': 'X19', 'F6': 'X9', 'F11': 'X20'}

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

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

< 10k

{'C1': '< 10k', 'C2': '> 10k'}

RandomForestClassifier

C2

Employee Attrition

There is disagreement about which label is acceptable for the case under consideration since the model is unsure which of the two labels is right. The confusion in the aforementioned classification may be attributable only to the effect of F18. F18 is by far the most influential variable, with a negative contribution that reduces the chance of label C2 being the correct label in the given case substantially; supporting the that case should be labelled as C1. Compared to the influence of F18, the remaining variables have a moderate to low effect on the classification decision made here for the case under consideration. F17, F28, and F24 are notable moderately key variables, with positive contributions boosting the likelihood of label C2. F7, F14, F9, F12, F30, F29, F4, F13, F11, and F8 are not among the features demonstrated to contribute to the classification above; since they have very insignificant impact on the model's conclusion here.

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

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

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

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

Stay

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

KNeighborsClassifier

C1

German Credit Evaluation

For the case under consideration, the probability of C2 being the correct label is only 12.50%, implying that there is an 87.50% chance that C1 is the true label. The decision above was arrived at mainly based on the values of the following variables F3, F2, and F1. Among these top variables, only F3 has a very strong positive impact on the model, increasing the likelihood of C1 prediction. The most important variables decreasing the prediction are F2 and F1 and the remaining two shifting the verdict away from C1 are F9 and F6. F5 and F4 are the lowest-ranked variables, less important to the prediction made here since they have a moderately low positive impact on the model.

{'C1': '87.50%', 'C2': '12.50%'}

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

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

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

Good Credit

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

DecisionTreeClassifier

C1

Hotel Satisfaction

With a high degree of confidence, close to 100 percent, the classifier's final label choice for the given case is C1 due to the predicted probability distribution between the class labels. Analysis of the attributions of the input features indicates that the most relevant features driving the classification above are F1, F9, F4, and F5, whereas F10 and F2 are shown to have little contribution to the decision. Furthermore, only four of the features have a negative influence, swinging the classifier decision in this case towards the C2 label and they are F9, F14, F10, and F2. However, except for F9, the contribution of the other negative features is very low when compared with the top positive features such as F4, F15, and F5.

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

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

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

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

satisfied

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

KNNClassifier

C2

Car Acceptability Valuation

Based on the values of the six input features, the model assigned the label C2 to the given case with a higher degree of confidence and according to the model used here, there is a near-zero chance that the label could be C1. Influencing the prediction assessment above are the top four features, F3, F4, and F2, whereas, the least significant feature here is F6. Among the input features, only two, F4 and F6, contradict the label assignment decision above since their values are shifting the label decision in the C1 direction. However, the joint attribution of these features is outweighed by the remaining four features, F3, F2, F5, and F1. This could explain why the model is very certain about the C2 prediction made for the case under consideration.

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

{'F3': 'persons', 'F4': 'buying', 'F2': 'lug_boot', 'F5': 'maint', 'F1': 'safety', 'F6': 'doors'}

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

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

Acceptable

{'C1': 'Unacceptable', 'C2': 'Acceptable'}

RandomForestClassifier

C1

Printer Sales

The most likely label for the given data is C1 and this decision is as the result of the variables passed to the classifier. F7, F17, F19, and F14 are the primary contributors to the aforementioned prediction output. F6, F5, F26, F15, F20, and F8, on the other hand, make insignificant contributions to the classifier labelling the given example. F25 and F21, as well as F13, F18, have a moderate influence on the label selection. The classifier's confidence in the label decision above might be explained away by comparing the greater positive attributions of F21, F25, F7, and F14 to the negative attributions of F13, F10, F19, F16, F17, and F22.

{'C2': '20.00%', 'C1': '80.00%'}

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

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

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

More

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

LogisticRegression

C1

Flight Price-Range Classification

Because the prediction algorithm outputs reveal that the likelihood of C1 being the correct label is equal to 93.02%; hence, there is only a little possibility that the true label for the provided data instance is either of the other labels, C3, C2, and C4. The variables F6, F12, F11, and F2 are the most crucial ones driving the label assignment conclusion above, whereas F1, F7, and F3 are the least vital ones. Taking into account the direction of effect of each input feature, as demonstrated by the attribution analysis, it is possible to deduce that the positive features driving the prediction upward towards C1 are F6, F4, F11, F5, F2, F12, and F7. The negative contributions of F8, F10, F3, F1, and F9 are ascribed to the marginal uncertainty in the expected output decision. When the predicted probabilities across the classes are considered, it is possible to infer that the combined positive contribution outranks the negative contributions; therefore, the algorithm is certain that C1 is the real label.

{'C1': '93.02%', 'C2': '6.97%', 'C4': '0.01%', 'C3': '0.0%'}

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

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

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

Low

{'C1': 'Low', 'C2': 'Moderate', 'C4': 'High', 'C3': 'Special'}

LogisticRegression

C1

Food Ordering Customer Churn Prediction

Based mainly on the values of the input variables F2, F7, F44, and F8, the predictor classifies the case as C1 with a 90.15% labelling confidence level, indicating that there is only a 9.85% probability that the right label could be C2. Variables that contribute positively to the prediction verdict include F2, F33, F39, and F8. The values of these variables increase the odds of the model labelling the given case as C1. On the other hand, F44, F7, F11, and F36 are the variables influencing the prediction decision in favour of C2 instead of C1. Simply put, the values of these negative variables contradict the label assigned here and finally, the model places little emphasis on the values of features such as F17, F21, F41, and F18 when determining the correct label in this instance, as they have nearly zero influence.

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

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

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

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

Go Away

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

LogisticRegression

C4

Air Quality Prediction

The classification output decision is based solely on the information supplied to the model and it predicts class C4 with a higher confidence level, equal to 94.10%, indicating the model is very confident that the correct label for the given case is not either class C3 or class C1 or class C2. The classification output decision with regards to the given case boils down to the values of the features F5, F4, F6, and F3, which are shown to have the most significant influence on the model. Among these relevant features, only F5, F3, and F4 have a positive impact, increasing the response towards labelling the case as C4. Conversely, the remaining ones, F6 and F1, have negative attributions, decreasing the odds of the assigned label. Finally, feature F2 has little impact on this prediction among the features since its value received little consideration from the model.

{'C3': '0.00%', 'C1': '0.53%', 'C4': '94.10%', 'C2': '5.37%'}

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

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

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

Cleaning

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

SVC

C2

German Credit Evaluation

The model predicts that this case is likely C2 with a confidence level equal to 66.80%, meaning there is a 33.20% chance that it could be C1 instead. According to the analysis for this case under consideration, the most relevant features considered by the model are F5, F2, F1, F8, and F7, however, the least relevant features are F9 and F4. The F1, F8, F7, and F6 can be regarded as positively supporting features given that they increase the model's response in favour of the prediction conclusion above. In contrast, the F5, F2, and F3 are the features supporting the prediction of the alternative or other class label C1. Even though only a small number of features support the prediction of C1, their collective or joint influence is enough to upset the joint influence of the other features, leading to the uncertainty of the C2 prediction.

{'C2': '66.80%', 'C1': '33.20%'}

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

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

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

Good Credit

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

SGDClassifier

C1

House Price Classification

Based on the values of the input variables resulting in the predicted likelihoods across the classes, the classification algorithm is confident that the right label for the provided data is C1. According to the algorithm, there is no possibility that C2 is the correct label. However, the attributions of F13, F11, F3, and F9 indicate that the correct label might be C2 rather than C1. The top four variables are F12, F7, F6, and F8, all of which have a positive influence on the algorithm's prediction output, hence confirming the C1 classification. This conclusion is further supported by the contributions of F2, F10, F1, F5, and F4, which are also positive variables.

{'C2': '0.0%', 'C1': '100.0%'}

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

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

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

High

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

KNeighborsClassifier

C2

Suspicious Bidding Identification

With a higher degree of confidence, the classifier assigns the label C2 due to the fact that there is a close to zero chance that C1 is the label. The confidence level with respect to this classification output is largely due to the strong positive influence of F2. However, decreasing the probability that C2 is the true label are the negative features F5, F7, F4, F3, F6, and F9. Furthermore, F1 and F8 also increase the likelihood of C2 being the true label. In conclusion, the joint impact of the negative features is very weak compared to the positive features, hence the strong driving force of the classifier to assign the chosen label, C2.

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

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

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

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

Normal

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

MLPClassifier

C2

Annual Income Earnings

According to the input variables, there is a 99.81% chance that C2 is the correct label for the given data instance, with a prediction probability of the alternative label, C1, equal to 0.19% which shows that there is little chance that C1 is the true label. F2, F9, and F6 are the top contributing features leading to the classification decision here. On the contrary, the F7, F12, and F1 are the least relevant features. The input features with moderate influence are F4, F10, F3, F5, F14, F8, and F13. Even though the different features have some level of influence on the classification, not all of them positively contribute. Actually, F3, F11, F13, and F12 have negative attributions, decreasing the classifier's response towards assigning C2; however, the joint influence of these features is outweighed by the positive attributions of F2, F9, F6, F4, and F10.

{'C2': '99.81%', 'C1': '0.19%'}

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

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

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

Under 50K

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

SVC

C2

Australian Credit Approval

With a confidence level equal to 81.43%, the classification algorithm labels the given data as C2, however, there is about an 18.57% chance that C1 could be the right label. The assignment of C2 to the given case is mainly based on the positive influence and contribution of input features F7, F1, and F6. Furthermore, the majority of the remaining input features have positive contributions, further increasing the predictability of label C2. F5, F10, F4, and F12 are the features with negative contributions, shifting the decision towards C1 instead of C2. Summarizing, comparing the attributions of the negative features to even those of the top three positive features explains why the algorithm is certain that C2 is the right label here.

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

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

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

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

Class 2

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

DNN

C1

Broadband Sevice Signup

The classification model employed here is very certain that the correct label is C1, implying that there is a near-zero chance that C2 is the label. The top six variables with the most influence on the prediction are all shifting the prediction in favour of C1. This might explain why the model is very certain about the C1 label and these top positive attributes are F14, F19, F29, F6, F38, and F20. F12, F36, F16, F27, and F26 all have moderately low-negative contributions, weakly swinging the direction of the model's decision towards C2. Finally, the decision to label the case as C1 is marginally supported by F2 and F33, whereas F13, F5, and F41 suggest that C2 could be the true label. In conclusion, the very high confidence level with regard to this prediction can be explained away by considering the very strong positive influence of F14, F19, F6, and F29.

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

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

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

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

Yes

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

KNeighborsClassifier

C2

Water Quality Classification

For the given case, the model generates the label C2 instead of C1, since C2 has a higher prediction likelihood than C1. According to the attribution graph shown, F7, and F9 are the most influential variables, resulting in the classification verdict above. F1, F4, and F6, on the other hand, are the least important variables considered by the model. F8, F2, F5, and F3 are shown to have a moderate influence on the classification made here. To sum up, with F1, F4, and F6 being the only variables contributing negatively, it is foreseeable why the model is quite certain that C1 is not the correct label for the given case.

{'C2': '87.50%', 'C1': '12.50%'}

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

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

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

Not Portable

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

DecisionTreeClassifier

C1

Concrete Strength Classification

Per the classifier, the most probable class with a very high confidence level is C1 mainly because the probability that C2 is the correct label is zero. From the attributions analysis, all the inputs are shown to contribute to or influence the above classification. The ranking of the features from the least important to the most important based on their degree of influence is as follows: F3, F5, F8, F2, F6, F7, F1, F4. Simply looking at the attributions of the input features, it is obvious why the classifier is very confident that C2 is not the correct label for the given All the features have positive contributions, resulting in a strong push towards C1.

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

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

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

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

Weak

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

KNeighborsClassifier

C3

Air Quality Prediction

The model predicted the C3 class for the test case with a very high degree of confidence. F2 is the only feature contributing against the prediction of the C3 class, while F3 and F5 contributed positively towards the prediction of C3. In decreasing order, F1, F6 and F4 were the three features with the least positive impact on the prediction of C3. Overall, given that only F2 has negative influence on the decision, it is not surprising to see the associated confidence level of the assigned label.

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

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

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

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

Presence of smoke

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

SVM_linear

C2

Mobile Price-Range Classification

Per the classification algorithm, the most probable class is C2 since the prediction probabilities indicate there is little to no chance that the correct label for the given data instance is any of the following classes: C1, C3, and C4. This labelling is primarily owing to the roles that the features F12, F13, and F3 performed. On the lower end of the spectrum are the input features F6, F17, F10, and F9, which are demonstrated to be less essential for this labelling assignment task. Finally, only F19 and F11 are features having a negative effect, reducing the likelihood of C2 being the accurate classification here.

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

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

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

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

r4

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

GradientBoostingClassifier

C1

Australian Credit Approval

The predicted label is C1 at a confidence level of 92.11%, insinuating that there is a 7.89% chance that the label could be C2. In this case, the feature with the most significant influence on the model's decision is F11, with a very strong positive contribution in support of the C1 prediction. The next set of features with moderately high impact is F4, F6, F10, F2, and F7. Among this set, only F7 and F4 have a negative influence in support of label C2. Finally, on the lower end, the values of F8, F5, and F12 are deemed less important by the model when labelling this case.

{'C2': '7.89%', 'C1': '92.11%'}

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

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

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

Class 2

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

SVM

C2

Customer Churn Modelling

Taking into account the values of the input features, the prediction model's output for the case under consideration is C2. Given that there is a 27.27% probability that it could be C1, this labelling decision is not 100.0% certain. For the case under consideration, the label assignment is mainly due to the values of F1, F8, F5, and F10. F10 is identified as the most important or relevant, while F9 is considered the least important, since its contribution to the model is only marginal. In terms of the influence direction of each feature, F10 and F8 have a very strong positive contribution, driving the prediction higher toward the C2 class followed by F1, F5, and F6 all with moderately positive influence, whereas F9 has a negligible positive impact on the model in this case. Finally, for this case, F4, F2, F7, and F3 all have a negative impact on the prediction verdict, however, their pull or influence is not enough to transfer predictions in the direction of another class label, C1.

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

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

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

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

Leave

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

SVMClassifier_liner

C2

Employee Attrition

The prediction output decision by the model is that the likelihood of label C2 is 94.15% and that of class C1 is only around 5.85%, meaning the model is certain that C2 is likely the true label for the given case. First of all, the classification is performed with negligible contributions from the variables F23, F30, F15, F8, and F12 since their attributions are very close to zero. However, examination or inspection of the attributions of the different variables reveals that F14, F25, F7, F13, and F16 are the highly influential ones driving the predicted probabilities across the classes. In addition, the decision about the correct label for this case is moderately influenced by the values of F19, F3, F26, F24, F22, F1, and F2. In terms of the direction of influence or contributions of the variables, F14, F7, F13, F3, and F26 are the top positive variables, encouraging the predicted output to be equal to C2. Pushing the decision towards the C2 label and further away from C1 are the contriutions of the variables F22, F1, F18, and F10. Finally, the 5.85% likelihood of C1 can be attributed to the negative contributions of the top negative variables F25, F16, F10, F24, and F2.

{'C2': '94.15%', 'C1': '5.85%'}

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

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

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

Stay

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

RandomForestClassifier

C1

Printer Sales

The most probable label, according to the classifier for the given data, is C1, which happens to have a higher predicted probability than that of C2. The major players in the above prediction output are F18, F10, F4, and F16. Conversely, F23, F12, F2, F15, F6, and F17 have negligible contributions when it comes to the classifier labelling the given case. Features such as F11, F14, F21, and F5 have a moderate influence on the decision. Comparing the stronger positive attributions of F10, F18, F14, and F21 to the negative attributions of F4, F16, F11, F19, F3, and F20 could explain why the classifier is quite confident in the label choice above.

{'C2': '20.00%', 'C1': '80.00%'}

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

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

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

More

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

AdaBoostClassifier

C4

Air Quality Prediction

The most likely label is C4 since there is a 30.83% chance it could be C1, a 35.74% chance it could be C4, and a 33.42% chance it could be C3. Therefore, the correct label is not C2, which the model is very certain about. The above decision is primarily controlled by the values F1, F6, F4, and F5 which are shown to have positive influences that support the model's classification judgement here. In contrast, the remaining features F2 and F3 negatively support the classification decision, decreasing the chances of C4 being the correct label. In view of the fact that the probability distributions across the classes, we can conclude that the model is very uncertain about which label is appropriate for the given data instance and the features F2 and F3 should be blamed for this.

{'C1': '30.83%', 'C4': '35.74%', 'C2': '0.00%', 'C3': '33.42%'}

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

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

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

Presence of smoke

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

KNeighborsClassifier

C1

Ethereum Fraud Detection

Because the prediction probability of C2 is equal to 0.0%, the presented case is labelled as C1 with a very high level of confidence. For this classification scenario, the input features that have the greatest influence on the end outcome are F22, F20, F37, and F10. F16, F23, F24, F5, and F19 have a mild impact. However, because F4, F27, F1, and F18 have insignificant attribution values, they have little influence on the model's judgement. Among the top features, F22, F20, F37, and F10, only F22 and F37 exhibit negative attributions that favour the least likely class, C2, whereas F20 and F10 positively support the model's classification result for the provided data. Finally, only F3 and F14 positively contribute to the model's decision among the remaining significant features: F14, F28, F29, F3, F9, F31, and F12.

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

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

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

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

Fraud

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

RandomForestClassifier

C4

Mobile Price-Range Classification

Between the four possible classes, the label for this case is predicted as C4, with a 73.08% likelihood that this is correct. With a likelihood of about 26.92%, the next probable label is shown to be C1. The prediction assessment above is mainly based on the values of the features F19, F17, F4, F12, and F13. The strongest impact came from F19, followed by F4, F17, F13, and F12. The collective contributions of the positive features F19, F17, F1, and F14 far outweigh the contributions of the negative attributes F4, F13, F12, and F15. Of the twenty attributes, majority of them are shown to have values pushing the prediction towards one of the three other possible classes and as such, it is surprising to see that the model is not 100% confident in the C4 prediction. On the grounds that the likelihood of C4 being correct is 73.08%, we can conclude that the model is quite confident with its final decision for the case under consideration.

{'C4': '73.08%', 'C1': '26.92%', 'C2': '0.00%', 'C3': '0.00%'}

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

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

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

r1

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