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

Area of Science:

• Sports Science
• Data Science
• Machine Learning

Background:

• Accurate athlete distance tracking is crucial for performance analysis and injury prevention.
• Global Positioning Systems (GPS) are often unsuitable for indoor sports.
• Developing alternative methods for distance prediction is essential.

Purpose of the Study:

• To evaluate machine learning techniques for predicting athlete distances in indoor sports without GPS.
• To compare the effectiveness of XGBoost Regressor, ElasticNet, Ridge, and Lasso Regression.
• To analyze prediction accuracy for total, sprinting, and running distances across different sports and genders.

Main Methods:

• Machine learning models including XGBoost Regressor, ElasticNet, Ridge, and Lasso Regression were employed.
• Athlete data from men's and women's soccer and lacrosse at the University of Notre Dame were used.
• Performance was evaluated using root-mean-square error, standard deviations, means, and 95% confidence intervals.

Main Results:

• XGBoost Regressor achieved the lowest root-mean-square error for total distance (97.962 ± 12.973).
• XGBoost also demonstrated superior performance in predicting sprint distance (91.616 ± 4.234) and running distance (137.103 ± 2.789).
• Performance varied across sports, genders, and contexts (game vs. practice).

Conclusions:

• Machine learning, particularly XGBoost Regressor, offers a viable solution for predicting athlete distances in indoor settings.
• Accurate distance prediction can inform strategies for optimizing team performance, preventing injuries, and managing player conditioning.
• Model selection should be tailored to specific sports, genders, and activity contexts for best results.