Preventing sports injuries in professional football enhances players' availability and performance. Machine learning, including artificial neural networks, provides the opportunity to build multivariable prognostic prediction models that can help develop personalised risk estimations for injury occurrence. This study aims to construct predictive methods for injury risk based on selected body composition and physical fitness in professional football players across four seasons. The study sample comprised of 121 professional football players (26.1 ± 4.2 years old) who represented this team over the study period. The investigation was conducted according to three methodological sets of assessments: (i) average season, (ii) initial performance, and (iii) 2 weeks before injury. The primary classifier to predict injury risk was the K-star, which had a sensitivity of 81%. This method used data from 2 weeks before injury, including age, experience, body composition and lower-limb explosive strength. In application, regular body composition assessments and lower-limb explosive strength seemed more accurate in predicting muscle injury occurrence in this specific professional club. Predictive models learned from real-world training data and injury information can significantly aid early detection of injury risk. Future research could be valuable if longitudinal studies with classification methods use external and internal workload data.
Martins et al. (Thu,) studied this question.