AI Optimizes Baseball Pitch Sequences, Boosting Season Performance

While baseball analytics often focuses on optimizing the final pitch in an at-bat, the broader impact of entire pitch sequences, including preceding setup pitches, on long-term season performance has been less explored. This research addresses this gap using counterfactual analyses on MLB Statcast data. A Transformer-based machine learning model was trained to predict the outcome of a pitch, specifically whether it would result in an in-play outcome or a swing-and-miss. Counterfactual pitch sequences were then generated by altering either the final pitch or a setup pitch, keeping other contextual factors constant. Optimal counterfactual selections were defined as those minimizing the predicted in-play probability. The study estimated the expected effects of these optimizations on pitchers' seasonal statistics, finding that optimizing both final and setup pitches could lead to substantial improvements, such as an increase of over 1.0 in K/9 (strikeouts per nine innings). The findings also offer practical insights into effective pitch locations, the importance of command, and expanding pitch selection options.