Customer Churn Prediction & Behavior Analysis

Overview

In this project, I analyzed customer data from an e-commerce platform to understand user behavior and predict customer churn. The main objective was to help businesses identify customers at risk of leaving and take proactive actions to improve retention.

Dataset Overview

The dataset includes 5,630 customer records with 20 features covering demographics, purchasing behavior, engagement, and satisfaction metrics.

Problem Statement

Customer churn directly impacts revenue and growth. The goal was to build a predictive model that identifies customers at risk of leaving, enabling businesses to take proactive retention measures.

Key Questions:

• Which customers are most likely to churn?
• What factors drive customer churn?
• How can we intervene to prevent churn?

Key Observations

Churn Patterns:

• Short Tenure: Customers with shorter tenure are more likely to churn
• Complaints: Users who submitted complaints have higher churn probability
• Low Engagement: Fewer orders and longer inactivity strongly linked to churn

Model Performance

Random Forest:

• Accuracy: 96.18%

XGBoost (Best Model):

• Accuracy: 96.98%
• Precision (Churn): 95%
• Recall (Churn): 86%

Key Drivers of Churn

• Tenure: Customer lifetime
• Customer complaints: Negative experiences
• Marital status: Single customers show different patterns
• Product preferences: Category-specific behavior
• Payment method: Payment preferences matter
• Days since last order: Inactivity indicator

Business Insights

• New customers require more attention to reduce early churn
• Improving customer support can significantly enhance retention
• Re-engagement campaigns should target inactive users
• Personalized marketing strategies using behavior patterns

Conclusion

This project highlights how machine learning can predict customer churn with high accuracy and generate actionable insights to improve retention, optimize engagement strategies, and increase revenue.