Kneita, Declan
This paper presents a transformer-based neural network model designed for predicting individual pitch outcomes and hit locations in Major League Baseball. Unlike traditional approaches, the model dynamically adapts to a batter’s recent performance and game context, enabling precise and actionable predictions. Its dual predictive capabilities allow for optimizing pitch sequencing and defensive alignments. The paper also introduces an optimal pitch selection framework, showcasing how the model can provide tailored recommendations for specific game scenarios, enhancing real-time decision-making in baseball.