About

This canvas implements a comprehensive machine learning pipeline to predict long-term user success on the Zerve platform by engineering 25 behavioral features, clustering users into 5 personas, and training ensemble models (Random Forest and Gradient Boosting) with SHAP explainability, survival analysis, and publication-quality visualizations. The workflow progresses from data loading and cleaning through feature engineering, success label definition, persona segmentation, ML modeling (achieving ~0.8 AUC), and generates actionable insights including event sequence heatmaps, conversion funnels, and geographic analysis to identify high-value power users.