Title: Learning, Visualizing, and Exploiting a Model for the Intrinsic Value of a Batted Ball

Abstract: We present an algorithm for learning the intrinsic value of a batted ball in baseball. This work addresses the fundamental problem of separating the value of a batted ball at contact from factors such as the defense, weather, and ballpark that can affect its observed outcome. The algorithm uses a Bayesian model to construct a continuous mapping from a vector of batted ball parameters to an intrinsic measure defined as the expected value of a linear weights representation for run value. A kernel method is used to build nonparametric estimates for the component probability density functions in Bayes theorem from a set of over one hundred thousand batted ball measurements recorded by the HITf/x system during the 2014 major league baseball (MLB) season. Cross-validation is used to determine the optimal vector of smoothing parameters for the density estimates. Properties of the mapping are visualized by considering reduced-dimension subsets of the batted ball parameter space. We use the mapping to derive statistics for intrinsic quality of contact for batters and pitchers which have the potential to improve the accuracy of player models and forecasting systems. We also show that the new approach leads to a simple automated measure of contact-adjusted defense and provides insight into the impact of environmental variables on batted balls.