About

This project is an end-to-end Customer Churn Prediction & Risk Scoring system built as part of the Zerve AI Hackathon – Proof of Power.

Using Zerve’s analytical canvas, I designed a complete production-style workflow that goes beyond model training and focuses on real business impact.

The pipeline covers:
• Data ingestion and cleaning of real-world telecom customer data
• Exploratory analysis to identify churn patterns across tenure, contract type, and services
• Feature engineering with aligned preprocessing for training and inference
• Interpretable Logistic Regression model training and validation
• Model evaluation using Accuracy, Recall, and ROC-AUC
• Feature importance analysis to explain key churn drivers
• A deployment-ready scoring logic that predicts churn probability and risk level for individual customers

The system enables proactive retention by identifying high-risk customers before churn occurs, making it suitable for real-world decision support and API-based deployment.