About

This canvas executes a comprehensive 30-day user retention analysis pipeline: starting with raw event data (409k rows, 3,622 users), it engineers 13 behavioral features from a 14-day early activity window, trains a Gradient Boosting model with SMOTE imbalance handling (PR-AUC improvement: +33%), performs SHAP-based attribution to identify key retention drivers (session depth, recency patterns, tool diversity), applies K-Means clustering (k=3) to segment users into behavioral archetypes (Drive-by at 0.7% retention vs. Habitual at 2.9%), conducts Markov sequence analysis on event flows, and synthesizes all findings into a product insights report with 8+ actionable recommendations for converting drive-by users into habitual retained customers.