PRISM: Predictive Risk and Injury Surveillance Model for Athlete Safety

Apoorva Rumale, Mahek Desai, Marjan Asadinia

Abstract: Injury prediction in athletes is a critical aspect of sports medicine, where early identification is essential for minimizing the risk of severe injuries and enhancing athletic performance. This study explores the use of machine learning (ML) techniques to assess injury risk by analyzing key physiological metrics such as heart rate, impact force, activity duration, and skin temperature. A dataset containing athlete performance data under different conditions is examined to identify the predictive relevance of these features. Traditional classification methods faced challenges due to the non-collinear nature of the data, whereas boosting techniques proved to be the most effective, delivering the highest prediction accuracy. This research evaluates various ML models, including boosting algorithms, neural networks, and conventional classifiers, comparing their predictive capabilities. Using AutoML, AutoGluon determined that the Weighted Ensemble model outperformed other approaches, achieving an accuracy of 81.17%. To complement the injury prediction model, this study also incorporates a Reinforcement Learning (RL)-based Training Recommendation Model, which dynamically adjusts training regimens based on real-time physiological and biomechanical feedback. By optimizing training intensity and workload distribution, the RL model helps minimize injury risk while ensuring an appropriate and effective training regimen. The recommendation system is designed to be flexible, allowing for continuous adaptation and improvement as more data becomes available. This integrated approach enhances informed decision-making for coaches, medical professionals, and athletes, facilitating proactive injury prevention while maintaining peak performance levels. By leveraging ML-driven injury prediction and RL-based training recommendations, this study presents a comprehensive, scalable, and adaptable framework for athlete management. The results emphasize the potential of artificial intelligence in enhancing sports science, reducing rehabilitation costs, and optimizing performance through personalized, data-driven interventions.