๐Ÿง  Customer Churn Prediction

End-to-end ML pipeline โ€” 3 models compared with ROC curves, feature importance & business insights  |  JK Chhabra

Python Scikit-learn Matplotlib Gradient Boosting
Dataset Size
5,000
Churn Rate
~32%
Best Model AUC
~0.93
CV-AUC Score
~0.92
Models Compared
3
Top Driver
Contract Type

Model Performance Comparison

Logistic Regression
AUC~0.82
CV-AUC~0.81
SpeedFast
Random Forest
AUC~0.91
CV-AUC~0.90
SpeedMedium
โœ… Best Model
Gradient Boosting
AUC~0.93
CV-AUC~0.92
SpeedThorough

ML Analytics Dashboard

Churn Prediction Dashboard
โญ View on GitHub ๐Ÿ“„ View Code

Business Insights

๐Ÿ”ด Highest Risk Segment
Month-to-Month customers with under 12 months tenure are ~65% likely to churn. Priority retention campaigns should target this group immediately.
๐Ÿ“Œ Contract Type is #1 Driver
Customers on Month-to-Month contracts churn at 3x the rate of annual contracts. Offering upgrade incentives at month 10โ€“11 can significantly reduce churn.
๐Ÿ“ž Support Calls Signal Risk
Customers with 3+ support calls per month are significantly more likely to churn. Proactive outreach after the 2nd call can reduce this risk.
๐Ÿ’ณ Payment Delays = Early Warning
Payment delays of 7+ days correlate strongly with upcoming churn. Use this as a real-time trigger for retention offers.