About

Analyzed 409,287 user events across 4,774 Zerve users
to identify which behaviors predict long-term retention.
Built a Random Forest churn prediction model achieving
98.5% accuracy using 23 engineered features.

Key finding: Users who engage heavily in their first 7
days almost never churn — early lifecycle engagement is
the #1 retention predictor (MI score: 0.578).

Features engineered using Zerve AI Agent across 4
dimensions: temporal dynamics, behavioral sequences,
interaction complexity, and anomaly signals.

Tools: Python, Scikit-learn, Pandas, Matplotlib,
Zerve AI Agent | Submitted for Zerve x HackerEarth
$10,000 Data Challenge