Predicting Injuries in Football Based on Data Collected from GPS-Based Wearable Sensors

Area of Science:

• Sports Science
• Biomechanics
• Data Science in Sports

Background:

• Increasing match intensity and frequency in professional football leads to higher physical player loads.
• Improper training models (overtraining or undertraining) elevate the risk of player injuries.

Purpose of the Study:

• To develop predictive decision-making models for non-contact lower body injuries in football.
• To create a tool for modeling player load and forecasting injury risk within upcoming microcycles.

Main Methods:

• Implemented rule-based, fuzzy rule-based, and machine learning (XGBoost) decision-making models.
• Utilized player external load data collected during training and matches.
• Evaluated model performance based on precision, recall, and F1-score.

Main Results:

• The XGBoost machine learning algorithm demonstrated superior performance in predicting injury risk.
• XGBoost achieved 92.4% precision, 96.5% recall, and a 94.4% F1-score.
• All implemented methods showed varying degrees of success in predicting injury risk based on player load parameters.

Conclusions:

• Machine learning, specifically the XGBoost algorithm, offers a highly effective approach for predicting non-contact lower body injuries in football.
• Data-driven predictive models can aid in optimizing training loads and mitigating injury risk for professional players.