About

Production‑ready churn prediction system built entirely in Zerve. The project loads tabular customer‑like data, cleans and engineers features, trains LightGBM and XGBoost models, and combines them into an ensemble to predict churn probability for each user. It includes full evaluation (AUC, F1, confusion matrix, ROC), business KPI calculations to estimate revenue at risk and ROI from retention campaigns, a reusable predict_churn function for real‑time scoring, and batch‑scoring logic suitable for APIs and scheduled jobs.