Injury Prediction and Risk Modelling in Team Sports Using Artificial Intelligence and Sensor-Based Monitoring: A...

Sports-related injuries remain a major challenge in team sports, with important consequences for athlete health, performance, and team success. Recent advances in artificial intelligence (AI) and sensor-based monitoring technologies have enabled the integration of large volumes of training, competition, and physiological data to support injury prediction and risk modelling. However, the literature is characterised by substantial methodological diversity, limiting the ability to draw consistent conclusions. Hence, this scoping review aimed to map the existing evidence on the use of AI and sensor-based monitoring technologies for injury prediction and risk modelling in team sports, and to identify key methodological trends and research gaps. The scoping review was conducted in accordance with the PRISMA-ScR guidelines. Systematic searches were performed in PubMed and Scopus. Eligible studies included team-sport athletes and applied AI or machine learning approaches to predict injury occurrence, injury risk, or related outcomes using data derived from wearable or monitoring systems. Data were charted on study characteristics, sports and competition level, data sources, modelling techniques, validation strategies, and performance metrics. The database search yielded 123 records (PubMed: n = 37; Scopus: n = 86). After screening and eligibility assessment, 11 studies met the inclusion criteria. Most studies focused on football and rugby and relied primarily on wearable-derived data, particularly GPS and inertial sensor outputs. Common predictors included external workload variables, training exposure, previous injury history, and, in some studies, wellness or physiological markers. A wide range of models was reported, including logistic regression, decision trees, random forests, support vector machines, and neural networks. Validation strategies and reported performance varied markedly, and external validation was rarely undertaken. Across the included studies, injury risk was most consistently associated with external workload metrics, previous injury history, and internal or physiological indicators of recovery and readiness. However, current models remain limited by heterogeneous methodologies, single-team datasets, and the lack of external validation. Future research should emphasise multimodal data integration and multi-centre validation to develop reliable, interpretable, and practically applicable AI-based injury prediction systems.