Machine learning-based injury risk prediction in male football players: A focused literature review

Abstract

Football injuries affect player availability, health, performance and financial burden. This focused literature review examined the use of machine learning (ML) to predict injury risk in male football players. English-language articles published between January 2021 and July 2026 were identified using PubMed and additional searches. 10 original that developed or assessed machine learning models with injury frequency, injury incidence or injury-risk categorisation as outputs were included. Predictors included preseason screening, GPS derived internal and external load, blood biomarkers and biomechanical measures with sample sizes ranging from 18 to 260 individuals. The models included neural networks, regularised regression models, tree-based approaches and support vector machines. Although several studies reported promising predictive performance, definitions of outcomes, prediction periods, class imbalance methods and validation methodologies were varied making meaningful comparisons among models difficult. Most models were developed on relatively small datasets, none reported independent external validation and limited calibration assessment. Overall, ML can be a tool to assist multidisciplinary injury surveillance, but it should not be the only factor in team selection or training decisions. This use of ML to injury-risk prediction in football is aligned with Sustainable Development Goal 3 (SDG 3), since it supports injury prevention, player wellness, and evidence-based sports medicine. Future study should aim for standardised outcomes, robust validation, multicentre datasets, calibration assessment and open reporting.