Machine Learning prediction of the expected performance of football player during training

Predicting athletes' performance is a fundamental task for the evaluation of their physical status, specific skills and quality of the training phase. This is even more important in team sports such as football, where the performance of the single athletes can significantly influence the whole team performance. Thanks to the improvements of modern sensors, which nowadays can be worn directly by the athletes within their vests without limiting them, it is possible to accurately measure several physiological parameters and movement indices that may be useful to make performance prediction. A growing number of studies is collecting these data and investigating relations among each other, in order to uncover possible correlations that can help in the prediction of sport performance. In this work, we used multivariate regression to attempt forecasting football players' performance during training sessions, starting from data about their movements and physiological parameters, extracted through a sensors array integrated in the vests worn by the athletes. The obtained results showed that specific combinations of physiological parameters can predict typical performance indicators, as acceleration and deceleration distances, as well as distance run under high-intensity effort, with an accuracy higher than 90%. The applied methodology can thus be used profitably by football teams' staff to monitor their players, giving it the possibility to make tactical decisions, design customized training sessions and orient market choices.