Using AI + Wearable Sensors to Predict and Prevent Sports Injuries in Real Time

Abstract

This study gives an AI-based on-body wireless sensor application in sports that detects and prevents sports injuries in real-time in competitive football, basketball, and track & field teams. The study included 60 athletes who were observed during 12 weeks and the parameters of the external load (total distance, high-speed running, acceleration/deceleration), as well as, the parameters of internal load (average heart rate, heart rate variability, session RPE), were quantified. Acute:Chronic Workload Ratio (Acute:Chronic Workload Ratio, ACWR), monotony, and strain levels were used as metrics of workload to determine the patterns of injury risk. Logistic Regression, Random Forest and LSTM Neural Networks were trained where LSTM performed best in predicting injury up to seven days in advance and performs best in ROC-AUC, recall and F1 score. The AI method was more effective than traditional methods of assessing the risks of injury, particularly in the early entry and lowering of false negativity rates. Wearables also offered actionable visual analytics, which made possible timely interventions. Their results emphasize the possibility of combining sensors-based monitoring and machine learning in order to improve athlete safety, training loads, and minimize injury occurrence.