Customer Segmentation using Decision Tree (ID3) on Online Retail Dataset

 

Experiment

Title

Customer Segmentation using Decision Tree (ID3) on Online Retail Dataset

---

🎯 Objective

• To segment customers based on purchasing behavior
• To implement Decision Tree using ID3 (entropy)
• To analyze:
  • Tree structure
  • Feature importance

---

📊 Dataset: Online Retail

Typical features:

• Quantity → Number of items purchased
• UnitPrice → Price per item
• Country → Customer location
• InvoiceDate → Purchase date
• CustomerID

---

🎯 Target (Segmentation Idea)

We create a simple segmentation:

👉 High Value Customer (1)
👉 Low Value Customer (0)

Based on:

$$\text{Total Spend} = Quantity \times UnitPrice$$

---

⚙️ Steps

1. Load dataset
2. Clean data
3. Create features
4. Define target
5. Train ID3 model
6. Visualize tree
7. Analyze feature importance

---

💻 Python Program

# -------------------------------
# 1. Import Libraries
# -------------------------------
import pandas as pd
import numpy as np
from sklearn.tree import DecisionTreeClassifier, plot_tree
from sklearn.preprocessing import LabelEncoder
import matplotlib.pyplot as plt

# -------------------------------
# 2. Load Dataset
# -------------------------------
# Download from:
# https://archive.ics.uci.edu/ml/datasets/Online+Retail

df = pd.read_excel("Online Retail.xlsx")

# -------------------------------
# 3. Data Preprocessing
# -------------------------------
# Remove missing values
df = df.dropna(subset=['CustomerID'])

# Remove negative/invalid values
df = df[(df['Quantity'] > 0) & (df['UnitPrice'] > 0)]

# Create Total Spend
df['TotalSpend'] = df['Quantity'] * df['UnitPrice']

# -------------------------------
# 4. Customer-Level Aggregation
# -------------------------------
customer_df = df.groupby('CustomerID').agg({
    'Quantity': 'sum',
    'UnitPrice': 'mean',
    'TotalSpend': 'sum',
    'Country': 'first'
}).reset_index()

# -------------------------------
# 5. Create Target Variable
# -------------------------------
# High Value if above median spend
threshold = customer_df['TotalSpend'].median()
customer_df['HighValue'] = (customer_df['TotalSpend'] > threshold).astype(int)

# -------------------------------
# 6. Encode Categorical Data
# -------------------------------
le = LabelEncoder()
customer_df['Country'] = le.fit_transform(customer_df['Country'])

# Features and target
X = customer_df[['Quantity', 'UnitPrice', 'Country']]
y = customer_df['HighValue']

# -------------------------------
# 7. Train Decision Tree (ID3)
# -------------------------------
model = DecisionTreeClassifier(criterion='entropy', max_depth=4)
model.fit(X, y)

# -------------------------------
# 8. Visualization
# -------------------------------
plt.figure(figsize=(20,8))

plot_tree(model,
          feature_names=X.columns,
          class_names=['Low','High'],
          filled=True)

plt.title("Customer Segmentation using Decision Tree (ID3)")
plt.show()

# -------------------------------
# 9. Feature Importance
# -------------------------------
importance = pd.Series(model.feature_importances_, index=X.columns)
print("\nFeature Importance:\n", importance.sort_values(ascending=False))

---

📈  Results

🔹 Tree Structure

• Root node often = Total spending related feature (Quantity / UnitPrice)
• Splits based on purchasing patterns

---

🔹 Feature Importance 

Feature Importance:
 Quantity     0.86527
UnitPrice    0.13473
Country      0.00000
dtype: float64

---

🌳 Interpretation of Tree

Example Logic:

If Quantity > threshold:
    → High Value Customer
Else:
    Check UnitPrice

---

🔍 Analysis

🔹 How Tree Helps Understand Customer Behavior

1. Spending Patterns

• High quantity → bulk buyers
• High price → premium buyers

---

2. Customer Segmentation

• Low spenders
• High spenders

---

3. Business Insights

Pattern	Insight
High Quantity	Wholesale customers
High UnitPrice	Premium buyers
Country influence	Regional behavior

---

📊 Key Insights

• Decision trees are:
  • Interpretable
  • Easy to visualize
• ID3 uses:
  • Entropy
  • Information gain

---

⚖️ Advantages

• Easy to explain
• Works well with categorical + numeric data

---

❌ Limitations

• Overfitting
• Sensitive to noise

---

🧪 Lab Tasks

Task 1

Change depth:

max_depth = 3, 5

---

Task 2

Add more features:

Invoice frequency

---

Task 3

Compare with:

criterion='gini'

---

Result

• Decision Tree (ID3) effectively segments customers
• Feature importance reveals:
  • Key buying factors
• Tree structure provides:
  • Clear business rules