A Data Mining Approach to Predict Non-Contact Injuries in Young Soccer Players

Abstract Predicting and avoiding an injury is a challenging task. By exploiting data mining techniques, this paper aims to identify existing relationships between modifiable and non-modifiable risk factors, with the final goal of predicting non-contact injuries. Twenty-three young soccer players were monitored during an entire season, with a total of fifty-seven non-contact injuries identified. Anthropometric data were collected, and the maturity offset was calculated for each player. To quantify internal training/match load and recovery status of the players, we daily employed the session-RPE method and the total quality recovery (TQR) scale. Cumulative workloads and the acute: chronic workload ratio (ACWR) were calculated. To explore the relationship between the various risk factors and the onset of non-contact injuries, we performed a classification tree analysis. The classification tree model exhibited an acceptable discrimination (AUC=0.76), after receiver operating characteristic curve (ROC) analysis. A low state of recovery, a rapid increase in the training load, cumulative workload, and maturity offset were recognized by the data mining algorithm as the most important injury risk factors.