Artificial Intelligence for Injury Risk Prediction in Competitive Sports A PRISMA-Compliant Systematic Review of Wearable, Biomechanical, and Training-Load Models

Abstract

Sport injuries continue to impact sport performance, health, and success of athletes, yet this has been a major challenge in competitive athletics. The development of artificial intelligence (AI), wearable sensors, and biomechanical analysis have provided new possibilities to assess the risk of injury predicting based on a large data set. The current systematic review sought to integrate evidence on AI-based injury predicting models using wearable sensor data, biomechanics data, and training-load data. PublMed, Scopus, Web of Science and IEEE Xplore were searched systematically according to PRISMA 2020 guidelines. Articles that were published between 2015 and 2025 that explored AI or machine-learning algorithms to predict injury in athletes were considered. Information on study design, sample, predictive algorithms, and accuracy measures were elicited. Among the 3124 records, 28 articles passed the inclusion criteria. The ontological algorithms that were mostly used are the machine-learning algorithms (Random Forest, Support Vectors, and neural networks). Wearable sensors, especially inertial measurement units, were commonly employed to make observations on the parameters of biomechanics and the variables describing the workload. Findings indicate that AI-based models demonstrate good performance of prediction, and reported accuracies in the model are between 75 and more than 90. The combination of wearable sensors data and training-load data showed a significant enhancement on prediction. Nevertheless, there are other issues such as methodological heterogeneity and small sample sizes. Injury prediction models based on AIs have a lot of potential in enhancing strategies of injury prevention in both sports science and monitoring of athletes.