About

This project implements a production-ready customer churn prediction system built entirely on the Zerve platform.

The system ingests telecom customer data, performs feature preprocessing and scaling, trains a Gradient Boosting classification model, and evaluates performance using business-relevant metrics including ROC-AUC, precision, recall, F1-score, and confusion matrices.

A hyperparameter optimization step is included to improve model effectiveness using cross-validated tuning, after which the best-performing model is selected for deployment.

The final model is deployed as a Zerve task that generates real-time churn predictions, probability scores, risk classification (LOW / MEDIUM / HIGH), confidence estimates, and actionable retention recommendations.

A lightweight web interface demonstrates real-world usage by allowing users to input customer profiles and visualize predictions and model performance. This system is designed to reflect a realistic, production-grade decision-support workflow for customer retention teams.