Churn Prediction System
Decision-focused churn ML with business-aware thresholds, ROI analysis and retention strategy.
Challenge
Churn models predict who leaves but never answer what to do about it — who to contact, how much to spend, and whether it pays. Default 0.5 thresholds either overspend retention or miss at-risk customers.
Solution
14 business hypotheses validated in EDA, 24 features across tenure, pricing and service buckets, then the decision layer: a CLV-based cost matrix (~$1,554 CLV, $50 intervention, 25% save rate), ROI-driven threshold sweeps and segment-specific thresholds. Random Forest at 0.20: 96.5% recall, $140K revenue saved for $33.9K spend — $106K net ROI across ~678 targeted customers, explored through a four-view Streamlit decision tool.
View GitHub Repo
