About

Comprehensive data science workflow analyzing 4,774 users to predict 30-day retention using behavioral features from first 7 days of activity. The canvas implements a complete machine learning pipeline: data loading → cleaning → feature engineering → retention label creation → exploratory data analysis → multiple model training (RF for 30-day retention + time-windowed Day 1/Day 3 models) → persona clustering → comprehensive visualizations (ROC curves, feature importance, retention curves, persona analysis) → production scoring system. Achieves 83.5% ROC-AUC on 30-day retention with key drivers being time span (46.5%), total events (15.3%), and active days (13.9%), while identifying behavioral personas ranging from low-engagement explorers (1% retention) to mid-engaged explorers (17% retention) with actionable product recommendations for improving multi-day engagement loops.