Dipartimento di Ingegneria informatica, automatica e gestionale

Over the course of a soccer season, intensified training periods may increase players' fatigue and impair recovery status. Therefore, understanding the internal and external load markers-related fatigue is crucial to optimize weekly training load and to improve players' performance. The aim of the current investigation was to adopt machine learning (ML) techniques to predict recovery status from a set of internal and external load parameters and identify which one had the greatest impact on players' recovery ability. Twenty-six adult soccer players were monitored for 6 months. Internal and external load parameters were daily collected. Players' recovery status was assessed through a modified 10-point total quality recovery (TQR) scale. Different ML algorithms were employed to predict players' recovery status in the subsequent training session (S-TQR). The goodness of the developed models was evaluated through root mean squared error (RMSE), mean absolute error (MAE), and Pearson's Correlation Coefficient (r). Among the different ML algorithms developed, random forest regression model produced the best performance (RMSE = 1.32, MAE = 1.04, r = 0.52). TQR, age of the players, total decelerations, average speed, and S-RPE recorded in the previous training were recognized by the model as the most relevant features. The ML approach allowed to predict the S-TQR score and to understand the most influential variables on players' recovery status.