Computing an Optimal Pitching Strategy in a Baseball At-Bat

The field of quantitative analytics has transformed the worldof sports over last decade. To date, these analytic ap-proaches are statistical at their core, characterizing what isand was, while using this information to drive decisionsabout do in future. However, as we often viewteam sports, such soccer, hockey, and baseball, pairwisewin-lose encounters, it seems natural model zero-sum games. We propose a for baseball at-bat,which is matchup between pitcher batter. Specifi-cally, novel encounter stochastic game, which goal batter geton base, an outcome aims prevent. valueof game on-base percentage (i.e., probabilitythat gets on base). In principle, gamecan be solved classical approaches. main techni-cal challenges lie predicting distribution pitch loca-tions function intention, distri-bution outcomes if decides swing pitch,and level patience particular batter.We address by proposing andbatter representations well deep neural networkarchitecture prediction. Our experiments usingKaggle data from 2015 2018 Major League Baseballseasons demonstrate efficacy proposed approach.