Bradley Bowen's GWD Thesis Presentation

Senior Bradley Bowen will present his thesis “Movement-Based Clustering of Pitches in Major League Baseball” for Graduation with Distinction on Wednesday, 15 April, at 9:30 AM EDT in Old Chem 211A. Bradley's thesis advisor is Dr. Alexander Fisher, Assistant Professor of the Practice of Statistical Science. His other departmental advisors are Dr. Jerry Reiter, Professor of Statistical Science, and Dr. Colin Rundel, Associate Professor of the Practice of Statistical Science. 

 

Bradley's majors are Statistical Science and Earth & Climate Science with a minor in Mathematics. 

 

Title: Movement-Based Clustering of Pitches in Major League Baseball

 

Abstract: Major League Baseball (MLB) organizations increasingly rely on data driven decisions as ball tracking technologies improve measurements of pitch movement, velocity, and batted ball characteristics. As part of this data collection, MLB assigns each thrown pitch a label such as a four-seam fastball, cutter, and changeup. However, these pitch classifications do not necessarily reflect how the pitch behaves. For example, two given pitchers may throw a “cutter” (as labeled by the MLB), but the movement profiles of each pitch may be very different. From a batter’s perspective, the movement behavior of the pitch matters more for swing decisions and contact quality than the technical label assigned to it. Therefore, we propose modeling the horizontal and vertical break of a pitcher’s throws using a Bayesian Gaussian mixture model to better characterize pitch behavior. We use previous season movement patterns across the league to write down informed semi-conjugate priors that facilitate an easy data-augmented Gibbs sampling scheme to approximate the posterior distribution of a pitcher’s throws. In this analysis, we examine data from 608 pitchers. We find that most pitchers exhibit three to four distinct movement pitch clusters as the model effectively collapses redundant or misclassified pitch types. As a downstream application, we use variables derived from the pitch movement clustering method to better predict batted ball attack direction, offering insight into the direction a batter will hit the ball.