Prediction of Subjective Fatigue in Professional Soccer Players: A Data-Driven Method to Optimize Training Approach to the Match

Abstract

In soccer, predicting players' fatigue experienced immediately before a training session or match can help design training programs and optimize performance. This study aimed to identify the most important predictors of daily and match-day fatigue in six Italian professional soccer teams during a competitive season using a framework of big data analytics. Every morning, the players rated fatigue, sleep quality, muscle soreness, stress, and mood. After each training session or match, the session Rating of Perceived Exertion was obtained and multiplied by duration to calculate the training load (TL). A framework of four machine learning models (Decision Tree classifier, XGBoost classifier, Random Forest Classifier, and Logistic regression) was trained and tested on 30.211 examples (one full season of six teams) to assess their ability to predict the players' match-day fatigue. The machine learning models accurately predicted the players' subjective fatigue (models' range accuracy 70-82%). Specifically, in the prediction of match-day fatigue, stress, and mood of the previous day were the most influential factors. Mediation analysis unveils the relationship between TL of the day before the match and the perception of match-day fatigue, also mediated by mood and muscle soreness. Sport scientists and coaches can apply this framework to simulate the effects of different training programs, thus maximizing players' readiness and mitigating potential drops in performance associated with match-day fatigue in a real-world scenario.

Keywords: Multidimensional approach; rating of perceived exertion; readiness; wellness.