About

This canvas implements a comprehensive ML/analytics workflow for Zerve user success prediction and behavioral analysis. It ingests raw event data (5,410 users, 100k+ events), engineers rich behavioral features, and trains three complementary predictive models:

1. User Archetype Discovery : K-means clustering (k=4) on 45 behavioral features produces four interpretable user personas (Power Coders, Collaborators, Casual Explorers, At-Risk Users) with per-cluster success rates and t-SNE visualization

2. Early-Signal Cluster Prediction : XGBoost/GBM trained on first-7-day behavioral signals to identify cluster membership at onboarding, enabling real-time intervention strategies

3. Success Likelihood Scoring : Full-history GBM with TreeSHAP explainability produces calibrated success probabilities (0-100 score tier) and identifies top predictive features

The workflow combines feature engineering (early-window, engagement velocity, feature adoption, collaboration metrics), rigorous model validation (5-fold CV, calibration, confusion matrices), SHAP analysis for interpretability, and per-cluster SHAP beeswarms for persona-specific insights. Outputs include 15+ visualizations (archetypes, SHAP beeswarms, success distributions, confusion matrices) and exportable probability/scoring tables for stakeholder reporting and onboarding automation.