Predicting Football Noncontact Injuries with Machine Learning
Author Information
Author(s): Freitas Diogo Nuno, Mostafa Sheikh Shanawaz, Caldeira Romualdo, Santos Francisco, Fermé Eduardo, Gouveia Élvio R., Morgado-Dias Fernando
Primary Institution: Interactive Technologies Institute (ITI/LARSyS), Funchal, Portugal
Hypothesis
Can machine learning techniques effectively predict noncontact injuries in professional football players using GPS data and player-specific parameters?
Conclusion
The study developed machine learning models that accurately predict noncontact injuries in professional football players, helping coaching staff identify high-risk players.
Supporting Evidence
- The models achieved a sensitivity of 71.43% and specificity of 74.19%.
- Key predictive factors included player position, session type, player load, velocity, and acceleration.
- The study combined GPS data with descriptive variables for improved injury prediction.
Takeaway
This study used computers to help predict when football players might get hurt without any contact, using data from their movements during games and practices.
Methodology
The study analyzed GPS data and player-specific parameters using machine learning models like Support Vector Machines, Feedforward Neural Networks, and Adaptive Boosting.
Limitations
The study was limited by the small number of injury events and did not include certain factors like muscle fatigue or specific exercises on rest days.
Participant Demographics
34 male professional football players from a Portuguese first-division team, mean age 26.27 years.
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website