3.1 Case Study Smart Retail Customer Segmentation Using Random Forest.ipynb

Kernel: Python 3 (ipykernel)

Random Forest is an ensemble learning algorithm that builds multiple Decision Trees and aggregates their predictions.

Key ideas:

• Uses Bootstrap Sampling (Bagging)

• Uses Feature Randomness

• Aggregates predictions using majority voting

Random Forest helps reduce:

• Overfitting

• Variance

A large retail chain wants to better understand its customers in order to personalize marketing campaigns and improve conversion rates.

The company collected behavioral data from 1000 customers including:

purchase frequency, website engagement, spending behavior, loyalty activity

The objective is to automatically classify customers into three segments:

Class Segment Meaning

• 0 Budget Buyers Highly price-sensitive customers

• 1 Regular Customers Moderate spending and engagement

• 2 Premium Customers High spending and high engagement

To solve this, we build a Random Forest classification model that predicts the customer segment based on behavioral features.

import numpy as np
import pandas as pd
import matplotlib.pyplot as plt

from sklearn.datasets import make_classification
from sklearn.model_selection import train_test_split
from sklearn.ensemble import RandomForestClassifier
from sklearn.metrics import classification_report, confusion_matrix, accuracy_score

retail_customer=pd.read_csv("retail_customer.csv")

retail_customer.head(3)

retail_customer.describe()

## Check Class distribition
retail_customer["Customer_Segment"].value_counts().plot(kind="bar")

plt.title("Customer Segment Distribution")
plt.xlabel("Segment")
plt.ylabel("Count")

plt.show()

## Prepare for ML
X = retail_customer.drop("Customer_Segment", axis=1)
y = retail_customer["Customer_Segment"]

X_train, X_test, y_train, y_test = train_test_split(
    X,
    y,
    test_size=0.2,
    random_state=42
)

rf = RandomForestClassifier(
    n_estimators=100,
    random_state=42
)

rf.fit(X_train, y_train)

y_pred = rf.predict(X_test)

print("Accuracy:", accuracy_score(y_test, y_pred))

print(classification_report(y_test, y_pred))

print(confusion_matrix(y_test, y_pred))

importance = rf.feature_importances_

plt.barh(columns, importance)

plt.title("Feature Importance")

plt.show()

Manual Gini Index Calculation (Using retail_customer Dataset)

Step 1: Select a Feature for Splitting

Assume the decision tree is considering a split based on:

• Annual_Spending

• Annual_Spending ≤ 0

• Annual_Spending > 0

Step 2: Count Class Distribution in Each Node

Assume after splitting the 1000 records we get:

Left Node (Annual_Spending ≤ 0)

Segment	Count
Budget_Buyer	180
Regular_Customer	90
Premium_Customer	30

Total = 300

Right Node (Annual_Spending > 0)

Segment	Count
Budget_Buyer	120
Regular_Customer	260
Premium_Customer	320

Total = 700

Step 3: Calculate Probabilities

Left Node

• P(Budget) = 180 / 300 = 0.6

• P(Regular) = 90 / 300 = 0.3

• P(Premium) = 30 / 300 = 0.1

Right Node

• P(Budget) = 120 / 700 = 0.171

• P(Regular) = 260 / 700 = 0.371

• P(Premium) = 320 / 700 = 0.457

Step 4: Apply Gini Formula

Step 5: Compute Gini for Left Node

Gini_left = 1 − (0.6² + 0.3² + 0.1²)

= 1 − (0.36 + 0.09 + 0.01)

= 1 − 0.46

= 0.54

Step 6: Compute Gini for Right Node

Gini_right = 1 − (0.171² + 0.371² + 0.457²)

= 1 − (0.029 + 0.138 + 0.209)

= 1 − 0.376

= 0.624

Step 7: Compute Weighted Gini for the Split

Step 8: Interpretation

Lower Gini means better separation.

Gini Value	Interpretation
0	Pure node
0–0.3	Very good split
0.3–0.6	Moderate split