Predicting Field-Sport Distances Without Global Positioning Systems in Indoor Play: A Comparative Study of Machine-Learning Techniques

A summary of the research:

This study addresses a challenge faced by field-based sports teams, like women's soccer, that often train indoors: Global Positioning Systems (GPS) don't work inside, making it difficult to track how far athletes run or sprint. This lack of data can hinder coaches' ability to plan effective training, manage player workload, and prevent injuries, as distance metrics are crucial for these decisions. The purpose of this research was to see if different machine learning computer programs could accurately predict these distances using other data available from athlete-worn devices, even when GPS signals aren't present.

To do this, researchers collected data from collegiate soccer and lacrosse athletes at the University of Notre Dame over several seasons. They evaluated four different machine learning models: XGBoost Regressor, ElasticNet Regression, Ridge Regression, and Lasso Regression. The main finding was that XGBoost Regressor consistently provided the most accurate predictions for total distance, sprinting distance, and running distance compared to the other models. This suggests that machine learning can fill the gap left by indoor GPS limitations, helping coaches and staff make data-driven decisions to support athlete performance and health year-round.

Here are three key practical takeaways from the article's results for women's soccer coaches and administrators:

• Implement XGBoost for Accurate Indoor Distance Tracking: XGBoost Regressor was identified as the most accurate machine learning model for predicting total, sprinting, and running distances in indoor settings. For women's soccer, this means coaches can use this technology to consistently monitor athlete workload and prepare training plans, even during off-season or bad weather indoor training periods, ensuring athletes are adequately prepared for competition.

• Utilize Predictive Models for Injury Prevention and Workload Management: Even with some variability, these predictive models provide sufficient "directional accuracy" to categorize training sessions (e.g., high intensity) and monitor training loads. For women's soccer, this ability to continually track distances is vital for injury mitigation programs, helping to identify and prevent potential non-contact and overuse injuries often linked to excessive training loads.

• Consider Simpler Models if Computational Demands are a Barrier: While XGBoost offers superior accuracy, it can be computationally complex and challenging for non-technical staff to interpret. ElasticNet Regression, though less accurate than XGBoost, offers a practical alternative by balancing strong prediction with greater interpretability. This means administrators can choose a model that best fits their team's resources and staff expertise, prioritizing either the highest accuracy or ease of use for effective athlete monitoring.

Authors: Casey J. Metoyer, Jonathon R. Lever, Alan Huebner, Holland A. Bill, Gabriel Tauro, Michael Labbe, David M. Smiley, Braden Kay, William Sovine, Jonathan D. Hauenstein, and John P. Wagle

You can read the entire article here.