E-Commerce Customer Churn Prediction

Cost-optimized machine learning with 98% ROC-AUC and 60% cost reduction through threshold optimization

Technology Stack

CatBoostXGBoostLightGBMSMOTESHAPStreamlitFastAPIScikit-learn

ML Frameworks

Deep Learning

NLP

Deployment

ROI

151%

Savings

$200K-$500K annually

Accuracy

98.1% ROC-AUC

Key Performance Indicators

Recall

99.3%

Cost Reduction

60% vs default threshold

Customers Saved

70 (from test set)

Business Impact Summary

Return on Investment	151%
Cost Savings	$200K-$500K annually
Model Accuracy	98.1% ROC-AUC
Recall	99.3%
Cost Reduction	60% vs default threshold
Customers Saved	70 (from test set)

Overview

Production-grade churn prediction with 98.1% ROC-AUC and 99.3% recall. Cost-sensitive learning with custom threshold optimization (0.14 vs default 0.5) reduces business costs by 60%. Only 1 churner missed out of 142 in test set.