advsp/maigurski-customer-personality-assignment1

Customer Personality Analysis – EDA Results

1. Project Goal

The goal of this project is to use numeric-focused Exploratory Data Analysis (EDA) on the Customer Personality Analysis dataset to understand:

• Which customer characteristics are associated with higher spending.
• How these characteristics differ between customers who responded to the last marketing campaign and those who did not.

The main outcome variable is:

• Response (0 = no, 1 = yes) – did the customer respond to the last marketing campaign?

The analysis is oriented around the question:

Which types of customers are more likely to respond to a marketing campaign, and how do they spend their money?

---

2. Dataset

• Name: Customer Personality Analysis
• Source: Kaggle – https://www.kaggle.com/datasets/imakash3011/customer-personality-analysis
• Local file used: marketing_campaign.csv
• Size: ~2,240 rows, ~29 columns (mostly numeric)

2.1 Main Feature Groups

• Demographics

  • Year_Birth, derived Age
  • Education
  • Marital_Status
  • Kidhome, Teenhome, derived Children_Total

• Spending (last 2 years)

  • MntWines, MntFruits, MntMeatProducts
  • MntFishProducts, MntSweetProducts, MntGoldProds
  • derived Total_Spending = sum of all product-related amounts

• Behavior / Engagement

  • Recency
  • NumDealsPurchases
  • NumWebPurchases
  • NumCatalogPurchases
  • NumStorePurchases
  • NumWebVisitsMonth

• Campaign / Feedback

  • AcceptedCmp1–AcceptedCmp5
  • Response
  • Complain
  • Z_CostContact, Z_Revenue

2.2 Derived Features Used in the Analysis

• Age = 2014 - Year_Birth
• Children_Total = Kidhome + Teenhome
• Total_Spending = sum of all product amounts
• Clipped features to reduce the influence of extreme values:
  • Income_clipped
  • Total_Spending_clipped

---

3. Data Preparation

3.1 Data Overview

• The dataset was loaded from marketing_campaign.csv (tab-separated).
• A basic overview was performed using:
  • df.shape
  • df.info()
  • df.head()

This confirmed that:

• The dataset is mostly numeric.
• A small amount of missing data exists (mainly in Income).
• No obvious structural issues appear.

3.2 Missing Values

• Missing values were checked with df.isnull().sum().
• The only variable with a meaningful number of missing values was Income.
• Missing Income values were imputed using the median:
  • Median is more robust than the mean in a highly skewed distribution.
• After imputation, Income had 0 missing values.
• Other variables had either 0 or negligible missing values and were left as-is.

3.3 Type Fixing

• Dt_Customer was converted from string to datetime using pd.to_datetime(..., dayfirst=True).
• This enables time-based calculations if needed (e.g. customer tenure).

3.4 Duplicates

• Duplicate rows were checked using df.duplicated().sum().
• Result: 0 duplicated rows → no rows were removed.

3.5 Outlier Handling

Outliers were examined mainly in:

• Income
• Total_Spending

Both were strongly right-skewed with very high extreme values.
Instead of removing customers, I used IQR-based clipping:

• Values below Q1 − 1.5 × IQR were set to the lower bound.
• Values above Q3 + 1.5 × IQR were set to the upper bound.

This produced the features:

• Income_clipped
• Total_Spending_clipped

and kept all customers in the dataset while reducing the impact of extremes.

---

4. Global Visualizations

Before asking specific research questions, I created a few global plots to understand the overall distribution and relationships.

4.1 Income Distribution (clipped)

• Income remains right-skewed, even after clipping.
• Most customers are concentrated in the lower-to-middle income range.
• A small group of customers still have relatively high incomes.

4.2 Campaign Response Distribution

• The vast majority of customers have Response = 0 (no response).
• Only a small minority responded to the last marketing campaign.
• This confirms that the target variable is imbalanced.

4.3 Correlation Heatmap – Selected Features

• There is a clear positive correlation between Income_clipped and Total_Spending_clipped.
• Web, catalog and store purchases are positively correlated with each other and with total spending.
• NumWebVisitsMonth is negatively correlated with purchases, suggesting that many visits without purchases may signal low engagement.
• Response has weaker but still informative correlations with spending and engagement features.

---

5. Research Questions, Plots, and Insights

Question 1 – Is there a positive relationship between income and total spending?

Insight:
There is a clear positive relationship between Income_clipped and Total_Spending_clipped.
Higher-income customers tend to spend more overall, even after clipping extreme values.
Not every high-income customer is a heavy spender, but heavy spenders are much more common in the higher income ranges.

---

Question 2 – Does total spending differ across age groups?

Insight:
The bar plot compares average Total_Spending_clipped between age groups.

• Middle-aged customers (around 40–59) show the highest average spending.
• Younger and older customers tend to spend less on average.

Age is therefore an important factor in understanding customer value.

---

Question 3 – Do children in the household affect spending patterns?

Insight:

• Households with fewer children (0–1) often have higher average spending.
• As the number of children increases, average spending per customer tends to decrease.

Family composition (number of children/teens at home) clearly influences how much customers spend.

---

Question 4 – Does marital status influence total spending?

Insight:

• The boxplot shows the distribution of Total_Spending_clipped across marital status categories.
• Some groups (for example, customers in stable relationships) contain more heavy spenders and have higher medians.
• Other groups have lower median spending and narrower distributions.

Marital status captures differences in household structure and financial behavior that affect consumption.

---

Question 5 – Do customers spend very different amounts on different product types?

Insight:

• Spending is not equally distributed across product categories.
• A large share of total spending is concentrated in wines and meat products.
• Other categories (fruits, fish, sweets, gold products) account for a smaller share of total revenue.

This shows that a few product lines drive most of the revenue, which is important for marketing and stock decisions.

---

Question 6 – Who responds to campaigns, and what characterizes them?

Insight:

The grouped bar plot compares average values of:

• Income_clipped
• Total_Spending_clipped
• Age
• Children_Total

for customers who responded to the last campaign vs those who did not.

Overall, responders tend to:

• Have higher income on average.
• Have higher total spending.
• Show somewhat different age and family profiles.

Responders are therefore more valuable and more engaged customers.

---

6. Overall Insights and Limitations

6.1 Key Insights

• There is a strong link between income and total spending: higher-income customers are more likely to be heavy spenders.
• Age, marital status, and number of children are all related to spending behavior and can be used for customer segmentation.
• Spending is concentrated in a few product categories, especially wines and meat products.
• Customers who respond to marketing campaigns tend to be higher-value customers, with higher income and spending levels.

These insights can be used to:

• Design targeted marketing campaigns.
• Focus on high-value customer segments.
• Improve cross-selling and upselling strategies across product categories.

6.2 Limitations

• The data is cross-sectional and comes from a single company, so results may not generalize to other contexts.
• Campaign responses and complaints are relatively rare, which limits statistical power.
• Outlier handling and imputation were based on robust but simple rules (median imputation and IQR clipping) and were not tuned for predictive modeling.

Despite these limitations, the EDA provides a clear picture of who the most valuable customers are, how they spend their money, and which groups are most likely to respond to future campaigns.

---

7. Presentation Video

You can watch the 2–3 minute walkthrough of this project here:
[[Watch the presentation video]https://www.youtube.com/watch?v=TfhvzoJg2ds