Customer Churn Analysis & Prediction 

Executive Summary

Customer churn represents one of the most critical challenges for SaaS companies, directly impacting revenue, growth, and customer lifetime value. Predicting churn risk accurately allows organizations to implement proactive retention strategies.

In this project, I analyzed customer behavior across multiple data sources, including subscriptions, product usage, support tickets, and churn events. These datasets sources were merged into a unified customer-level dataset to enable comprehensive analysis.

Several predictive models were developed and compared, including Logistic Regression, Random Forest, and XGBoost. Feature engineering, multicollinearity analysis, Recursive Feature Eliminations (RFE), and threshold optimization were applied to improve performance and interpretability.

The final XGBoost model achieved:

• 96.7% recall for churn prediction

• Perfect Specificity (1.000)

• ROC AUC close to 0.99

These results demonstrate strong predictive performance, enabling accurate identification of at-risk customers while minimizing false positive.

Alongside predictive modeling, the project also identified key behavorial drivers of churn, providing actionable insights to support data-driven retention strategies.

Key Business Insights

• Poor customer support experience significantly increases churn risk

• Higher product usage reduces churn probability

• Monthly billing customers churn more frequently than annual subsribers

• Customer behavior patterns vary by industry and country

Tools & Technologies

• Pyhton

• Pandas & NumPy

• Scikit-Learn

• XGBoost

• Matplotlib & Seaborn

• Jupyter Notebook

Model Performance

Multiple models were developed and evaluated:

• Logistic Regression for interpretability

• Random Forest for non-linear relationships

• XGBoost for final predictive performance

The XGBoost model achieved the best overall performance, maximizing churn recall while maintaining high specificity.

Next Steps

• Cost-benefit analysis of churn intervention strategies

• Model monitoring and retraining

• Deployment as a scoring pipeline

• Integration with analytics dashboards