About

A complete end-to-end ML pipeline built on 3.5M+ real Zerve product events to predict which free users will upgrade to a paid plan.
What this notebook covers:

Timestamp-based leakage prevention (0 leaked rows verified)
22 behavioural features engineered from pre-upgrade events only
Random Forest classifier (ROC-AUC 0.853, 2.8× best segment lift)
4-stage user funnel: Onboarded → Activated → Power User → Converted
Segment analysis identifying users converting at 2.8× base rate
Live Gradio app for real-time upgrade scoring

Key finding: Users don't upgrade after a long journey — they upgrade when they hit a resource limit. Real-time triggers beat weekly batch scoring for this product.
Built for: Zerve × ODSC AI Datathon
Model: Random Forest | ROC-AUC: 0.853 | Base rate: 5.3% | Users: 17,485