Transforming Pitching: A Data-Driven Approach to Mechanics and Coaching Using Deep Learning

Publication Date

Spring 4-20-2025

Presentation Length

Poster/Gallery presentation

College

College of Sciences & Mathematics

Department

Math and Computer Science, Department of

Student Level

Undergraduate

SPARK Category

Knowledge

Faculty Advisor

Christina Davis

WELL Core Type

Service Wellness

Metadata/Fulltext

Fulltext

SPARK Session

10:15-11:45

Presentation Type

Poster

Summary

This project explores whether transformer-based deep learning architectures can
effectively capture the complex biomechanics of baseball pitching and translate those insights into concrete coaching recommendations. By drawing on motion capture (joint angles, velocities, body landmarks, ground reaction forces, and energy flow) the system learns to predict pitch velocity while pinpointing which elements of a pitcher’s motion most influence performance. Beyond quantitative results, the transformer framework offers interpretable outputs that categorize pitchers based on mechanical efficiency and overall force generation. These insights support more nuanced, athlete-specific guidance, such as pinpointing technique refinements or recommending targeted strength training. By
highlighting which biomechanical factors limit a pitcher’s performance, the project
demonstrates the capacity of transformer-based models to move beyond raw predictions and deliver actionable feedback. Ultimately, this work shows how leveraging advanced sequence modeling in sports analytics can aid coaches, trainers, and athletes in optimizing pitching mechanics.

This document is currently not available here.

Share

COinS