Green Light: Using Logistic Regression to Predict When a Stolen Base Attempt Will Succeed

Authors

• Julio D. Karruli, University of Central FloridaFollow

Streaming Media

Files

Course Code

STA

Course Number

4504

Faculty/Instructor

Instructor Nathaniel Simone

Faculty/Instructor Email

nathaniel.simone@ucf.edu

Abstract, Summary, or Creative Statement

This project aims to predict the probability of a stolen base attempt of second base in the 2023 Major League Baseball (MLB) season using Logistic Regression. Using play-by-play event data from Retrosheet and player/pitch tracking data from Baseball Savant's Statcast system, this research investigates 2,800 stolen base attempts of second base with a baserunner only on first base to decide which factors are most important to a successful steal of second base using characteristics such as sprint speed, pop time, distance of lead, and pitch metrics. The objective of this research is to aid MLB coaches and front offices in evaluating baserunning ability that goes beyond traditional scouting methods. Additionally, given the game situation, for example, run differential and inning, we can optimize when a stolen base is most effective and necessary, compared to when it is a poor choice.  Model selection was performed by using backward selection and the Akaike Information Criterion (AIC) as our selection criterion. We will evaluate the final model by using accuracy metrics such as sensitivity and specificity, as well as building a Receiver Operating Characteristic (ROC) curve and analyzing the Area Under the Curve (AUC) to assess predictive accuracy.

Keywords

Stolen Bases; MLB; Logistic Regression; Baseball

Recommended Citation

Karruli, Julio D., "Green Light: Using Logistic Regression to Predict When a Stolen Base Attempt Will Succeed" (2026). High Impact Practices Student Showcase Spring 2026. 16.
https://stars.library.ucf.edu/hip-2026spring/16