Projections translate current pitch-level skills into expected future outcomes. The goal is a stable baseline expectation — not predicting hot streaks or slumps, but identifying where a pitcher's underlying skills point.
Next-season projections for ERA, WHIP, K%, BB%, GB%, and IP. ElasticNet and XGBoost are compared for every target, with separate starter and reliever models trained on pitcher-season transitions using Statcast, FanGraphs, and Stockyard grade features.
Direct predictions of next-season skill rates and innings, each from whichever model wins leave-one-year-out validation.
Benchmarked against Marcel-style baselines
Derived from projected K%, BB%, and GB% with a fitted SIERA-style formula, so the headline numbers stay consistent with the component projections.
