Sports Analytics
Statistical models and ML research in baseball and basketball. Projects: multi-model MLB pitcher injury prediction (Pitcher Injury Risk+), batting average skill isolation (BAA), and NBA shot-value estimation with SHAP interpretability.
Pitcher Injury-Risk Model
The best honest model I built predicts MLB pitcher injury at walk-forward AUC 0.574, barely above chance, and that number is the most important finding. Across 3,249 pitchers and 205,911 pitcher-game rows (Statcast 2015–2024, 82 features), I mapped where the domain ceiling sits and which signals stay useful below it: calibrated probabilities, a season-stable Injury Risk+ score, and a validated velocity dose-response.
Batting Average Ability (BAA)
A same-season decomposition that splits each hitter's batting average into a skill component and a luck component, then measures how much is skill: an ICC of 24.7%, built with mixed-effects modeling over 4,374 MLB player-seasons.
NBA Shot-Value Model
Measured what NBA shot context alone is worth by deliberately excluding shooter identity. Ran four model classes (LR, DT, RF, XGBoost) on raw vs. engineered features to test whether the ~0.64 AUC was an information constraint, then used calibrated probabilities to build an expected-value curve by distance and contest pressure.