Marketing Campaigns Analysis with Python

Problem scenario

Marketing mix stands as a widely utilized concept in the execution of marketing strategies. It encompasses various facets within a comprehensive marketing plan, with a central focus on the four Ps of marketing: product, price, place, and promotion.

Missing Value Imputation

There are missing income values for some customers. To address this, we assume customers with similar education and marital status tend to have comparable yearly incomes. We impute missing income values using the group mean based on these two variables. It's also necessary to ensure that `Education` and `Marital_Status` categories are cleaned before imputation.

    # Checking Education and Marital_Status unique values for cleaning
    print(data['Education'].unique(), data['Marital_Status'].unique())
    
    # Output should look like this:
    # array(['Graduation', 'PhD', '2n Cycle', 'Master', 'Basic'], dtype=object),
    # array(['Divorced', 'Single', 'Married', 'Together', 'Widow', 'YOLO',
    #        'Alone', 'Absurd'], dtype=object)
    
    # Imputing missing Income values based on Education and Marital_Status group mean
    data['Income'] = data.groupby(['Education', 'Marital_Status'])['Income'].transform(lambda x: x.fillna(x.mean()))
    

Feature Engineering

This step derives new features that capture key behavioral and demographic patterns:

• total_children: sum of children at home (Kidhome + Teenhome)
• Age: derived from the customer's year of birth
• total_spending: sum of all product category expenditures
• total_purchases: total purchases across all purchase channels

  # Create total number of children
  data['total_children'] = data['Kidhome'] + data['Teenhome']

  # Derive customer's age
  data['Age'] = 2025 - data['Year_Birth']

  # Calculate total spending across product categories
  spending_cols = ['MntWines', 'MntFruits', 'MntMeatProducts', 
                  'MntFishProducts', 'MntSweetProducts', 'MntGoldProds']
  data['total_spending'] = data[spending_cols].sum(axis=1)

  # Calculate total purchases across channels
  purchase_cols = ['NumWebPurchases', 'NumCatalogPurchases', 'NumStorePurchases']
  data['total_purchases'] = data[purchase_cols].sum(axis=1)
  

Exploratory Data Analysis (Income) & Outlier Treatment

To better understand the distribution of income values and identify potential anomalies, we use boxplots and histograms. Outliers are detected using the interquartile range (IQR) method. Income values that fall outside of the IQR range are capped to the boundary limits to reduce skewness and improve robustness.

  # Calculate IQR bounds
  Q1 = data['Income'].quantile(0.25)
  Q3 = data['Income'].quantile(0.75)
  IQR = Q3 - Q1
  lower = Q1 - 1.5 * IQR
  upper = Q3 + 1.5 * IQR

  # Save original income values
  data['Income_Original'] = data['Income']

  # Apply outlier capping
  data['Income'] = np.where(data['Income'] > upper, upper,
                  np.where(data['Income'] < lower, lower, data['Income']))
  

Outlier Treatment: Age

Similar to income, age can also contain extreme values that may distort the analysis. We apply outlier treatment using the IQR method to cap ages outside a reasonable range. This ensures more stable statistical modeling and visualization.

  # Save original Age values
  data['Age_Original'] = data['Age']

  # Calculate IQR for Age
  Q1 = data['Age'].quantile(0.25)
  Q3 = data['Age'].quantile(0.75)
  IQR = Q3 - Q1
  lower = Q1 - 1.5 * IQR
  upper = Q3 + 1.5 * IQR

  # Apply outlier capping
  data['Age'] = np.where(data['Age'] > upper, upper,
              np.where(data['Age'] < lower, lower, data['Age']))
  

To better illustrate the impact of IQR-based outlier treatment on the Age variable, we display separate histograms before and after capping. Notice that extremely high values (e.g., age 100+) are removed in the post-treatment view.

Feature Encoding & Correlation

We apply feature engineering to prepare the data for modeling. First, we encode Education using an ordinal scale. Then, we group unusual values in Marital_Status under "Other", and apply one-hot encoding to convert this categorical field into binary indicators. Finally, we generate a correlation matrix to understand how features relate to each other.

  # Ordinal encoding for Education
  edu_order = {'Basic': 0, '2n Cycle': 1, 'Graduation': 2, 'Master': 3, 'PhD': 4}
  data['Education_encoded'] = data['Education'].map(edu_order)

  # Group rare marital status categories
  data['Marital_Status_Clean'] = data['Marital_Status'].replace({
      'Absurd': 'Other',
      'YOLO': 'Other',
      'Alone': 'Other'
  })

  # One-hot encoding
  data = pd.get_dummies(data, columns=['Marital_Status_Clean'], prefix='Marital')
  

Focused Correlation Heatmap

Rather than examining all 30+ variables at once, we focus on a curated set of key features such as age, income, education level, and purchase behavior. This cleaner correlation matrix makes it easier to identify important relationships, such as how income relates to total spending, or how age correlates with recency or number of children.

  # Select features most relevant to behavior and segmentation
  selected_features = [
      "Age", "Income", "Education_encoded", "total_spending",
      "total_purchases", "total_children", "Recency", "Complain"
  ]

  # Compute correlation matrix for just these variables
  corr = data[selected_features].corr()
  

Hypothesis A: Older Individuals Prefer In-Store Shopping

We test the idea that older individuals may have lower technological proficiency and therefore prefer to shop in-store. We analyze the correlation between age and the number of store purchases, and compare purchase behavior across age groups.

# Correlation between Age and NumStorePurchases
data['AgeGroup'] = pd.cut(data['Age'], bins=[18, 30, 45, 60, 75, 100], labels=['18–30', '31–45', '46–60', '61–75', '76+'])
correlation = data[['Age', 'NumStorePurchases']].corr()
  

Hypothesis B: Parents prefer web purchases

Individuals with children may face time constraints and be more price-sensitive. To explore whether they prefer online shopping, we grouped customers by number of children and compared their web purchase activity. This is visualized using a boxplot and scatterplot, and evaluated statistically using correlation.

  # Create total_children feature
  df['total_children'] = df['Kidhome'] + df['Teenhome']
  
  # Filter relevant fields
  df_b = df[['total_children', 'NumWebPurchases']].dropna()
  
  # Correlation between children and web purchases
  corr = df_b.corr().iloc[0, 1]
  print(f"Correlation: {corr:.4f}")