NBA Elo Analysis

Goal: Compare refreshed Elo variants and modern ML with FiveThirtyEight’s NBA forecast.

Data:~16 k box-score games (1981-2015).

Plan:
• Engineer pace, shooting, rebounding, and defense differentials.
• Train six models: Moving-Average, Classic Elo, Elo-LASSO, XGBoost, Logistic, plus 538 Elo as external benchmark.
• 70/30 chronological split; 5-fold CV on train. Metrics: accuracy, ROC-AUC, precision, recall, confusion matrices.

Deliverables:
1. Ranking the models
2. Short-list of important factors.
3. Public GitHub repo.