Projecting MLB Pitcher Performance with Pitch Grades
Open Access
Author:
Zerbe, Malcolm
Area of Honors:
Statistics
Degree:
Bachelor of Science
Document Type:
Thesis
Thesis Supervisors:
Matt Slifko, Thesis Supervisor Andrew John Wiesner, Thesis Honors Advisor
Keywords:
Baseball MLB Machine Learning
Abstract:
Baseball has historically been a game deeply rooted in statistics and numerical performance evaluation. From the batting averages of Ted Williams and home run totals of Hank Aaron to the earned run averages of Clayton Kershaw and on-base percentages of Juan Soto, the way we understand and interpret player performance has continuously evolved. Today, more advanced methods of quantifying player performance and the rise of advanced data mining have ushered in a new era of baseball analytics. This thesis ventures into the forefront of this new era, focusing on the predictive analysis of starting pitcher performance in Major League Baseball. With the advent of model-based pitch grades estimating both the quality of the physical characteristics of pitches and of the location of pitches, there is an opportunity to enhance our predictive capabilities. This study aims to bridge traditional baseball wisdom with the cutting-edge techniques of machine learning, to see if these new metrics can improve the accuracy of future-season starting pitcher performance predictions. By building and comparing projection models both with and without these new metrics, this research seeks to unveil their true predictive power and potentially redefine how we project pitching performance.