[Paper Review] Performance Prediction in Major League Baseball by Long Short-Term Memory Networks

Hsuan-Cheng Sun, Tse-Yu Lin+1 more|arXiv (Cornell University)|Jun 20, 2022

Sports Analytics and Performance4 citations

TL;DR

This paper proposes using Long Short-Term Memory (LSTM) networks to predict home run performance in Major League Baseball (MLB) based on historical player statistics. The study demonstrates that LSTM models outperform traditional machine learning models and the widely used sZymborski Projection System (ZiPS), achieving lower MAE and RMSE, especially in 2019, and providing more accurate predictions within small error intervals.

ABSTRACT

Player performance prediction is a serious problem in every sport since it brings valuable future information for managers to make important decisions. In baseball industries, there already existed variable prediction systems and many types of researches that attempt to provide accurate predictions and help domain users. However, it is a lack of studies about the predicting method or systems based on deep learning. Deep learning models had proven to be the greatest solutions in different fields nowadays, so we believe they could be tried and applied to the prediction problem in baseball. Hence, the predicting abilities of deep learning models are set to be our research problem in this paper. As a beginning, we select numbers of home runs as the target because it is one of the most critical indexes to understand the power and the talent of baseball hitters. Moreover, we use the sequential model Long Short-Term Memory as our main method to solve the home run prediction problem in Major League Baseball. We compare models' ability with several machine learning models and a widely used baseball projection system, sZymborski Projection System. Our results show that Long Short-Term Memory has better performance than others and has the ability to make more exact predictions. We conclude that Long Short-Term Memory is a feasible way for performance prediction problems in baseball and could bring valuable information to fit users' needs.

Motivation & Objective

• To explore the feasibility of deep learning, specifically LSTM networks, for predicting future player performance in MLB.
• To evaluate LSTM’s predictive accuracy for home runs—a key indicator of hitter power and talent—against established models.
• To provide actionable insights for baseball decision-makers by developing a more accurate prediction system based on sequential player statistics.
• To establish a new baseline for performance prediction in baseball using deep learning, addressing a gap in current literature.

Proposed method

• The study uses a sequential modeling approach with Long Short-Term Memory (LSTM) networks to process time-series player statistics.
• Input features include 20 historical performance metrics such as at-bats, hits, home runs, and walks, aggregated over multiple seasons.
• Five different LSTM architectures are trained and compared, with hyperparameter tuning to optimize prediction performance.
• Model performance is evaluated using MAE (Mean Absolute Error) and RMSE (Root Mean Square Error) on test data from 2018 and 2019.
• The predictions are benchmarked against a widely used projection system (sZymborski Projection System, or ZiPS) and traditional machine learning models like linear regression and random forest.
• The model assumes constant player height and weight across seasons due to data limitations, and does not include advanced metrics like hard-hit rate or BABIP.

Experimental results

Research questions

• RQ1 Can LSTM networks effectively predict future home run production in MLB players using historical performance data?
• RQ2 How does the predictive performance of LSTM compare to traditional machine learning models and the established ZiPS projection system?
• RQ3 What is the impact of sequential modeling on prediction accuracy for player performance in sports?
• RQ4 Why do high-home-run players (e.g., >30 HRs) prove difficult to predict, and can LSTM models address this challenge?

Key findings

• LSTM models achieved the lowest MAE and RMSE on the 2019 test set, outperforming all other models, including linear regression and ZiPS.
• In 2018, linear regression had similar low RMSE and MAE to LSTM, but LSTM significantly outperformed all others in 2019.
• LSTM models made the most accurate predictions within small error intervals, while other models showed large deviations between predicted and actual home run counts.
• The study confirms that deep learning, particularly LSTM, is a robust and effective method for sequential performance prediction in baseball.
• The results suggest that LSTM can serve as a new baseline for future research and practical applications in player performance forecasting.
• High-home-run players (>30 HRs) remain difficult to predict, indicating a need for further investigation into model architecture and feature engineering.