A Narrative Review for a Machine Learning Application in Sports: An Example Based on Injury Forecasting in Soccer.

Area of Science:

• Sports Science
• Data Science
• Computational Science

Background:

• The application of machine learning (ML) in sports science has surged, particularly for injury forecasting and athlete performance prediction.
• Existing research presents numerous ML techniques, necessitating a structured approach to model development and validation.

Purpose of the Study:

• To provide a comprehensive guideline for the correct training, validation, and testing of ML models in sports science.
• To identify and address potential strengths and limitations across all stages of ML model development.
• To minimize errors and prevent misleading results in sports event prediction models.

Main Methods:

• This narrative review synthesizes current literature on ML methodologies in sports science.
• It outlines best practices for data preprocessing, feature selection, and dataset splitting for time-series analysis.
• The review emphasizes the importance of model interpretation for both black-box and white-box models.

Main Results:

• The review details features applicable to injury forecasting and various time-series preprocessing techniques.
• It provides guidance on appropriate dataset splitting strategies for robust model training and testing.
• The importance of explaining the decision-making process of ML models is underscored.

Conclusions:

• Adhering to rigorous training, validation, and testing protocols is crucial for reliable ML models in sports science.
• Understanding model limitations and interpretation is key to avoiding erroneous predictions in athlete performance and injury forecasting.
• This guideline aims to enhance the accuracy and trustworthiness of ML applications in the sports domain.