Prédiction des blessures des ischiojambiers en football à l'aide d'apprentissage automatique : étude préliminaire sur 284 footballeurs

The hamstring injury is the number one injury diagnosis in football. One of the strategies for hamstring injury prevention is the identification of high-risk athletes. In this article, we implemented machine learning algorithms for injury risk estimation in 284 male footballers playing in 16 professional or semi-professional football teams from 3 countries. The predictors (input data) of the algorithms consisted of an athlete's baseline dataset, including hamstring injury history in the previous season and data measured during a maximum sprint of 30 m. The output data, binary, was the occurrence of hamstring injury. The three models used, logistic regression, random forests and AdaBoost, were compared to a dummy classifier. The results showed that it is possible, to a certain extent, to predict the occurrence of injury with these models. The comparison with the dummy classifier, when considering a set of metrics including F1-score, showed the interest of the three models used. Additionally, the relative importance of predictors can be measured, which can aid in understanding the predominant factors influencing injury. These results suggest avenues for hamstring injury prevention strategies.