Are cognitive functions capable of predicting the overall effectiveness and the contributions of athletes to the team in small-sided soccer games? A machine learning approach

Author(s)

Rafael Luiz Martins Monteiro

• Document type: Master's Dissertation
• Press: Ribeirão Preto.
• Institution: Universidade de São Paulo (USP). Faculdade de Medicina de Ribeirão Preto (PCARP/BC)
• Defense date: 2024-10-16

Examining board members

Paulo Roberto Pereira Santiago; Joaquim Cezar Felipe; Paulo Mazzoncini de Azevedo Marques

• Advisor: Paulo Roberto Pereira Santiago

Abstract

Soccer is a dynamic team sport characterized by high cognitive demand, where quick and efficient decision-making is necessary. The present study aimed to explore how cognitive function data are related to the performance of youth high-level soccer players on different demands of small-sided games using machine learning algorithms to predict the performance level of the athletes and describe the most important cognitive functions for each game demand prediction. Forty-four male athletes (16.51±0.57 years old) from the U-17 category of two soccer teams participated in the study. For on-field performance evaluation, a small-sided multiple games protocol was implemented, with 3 vs 3 player matches without a goalkeeper. Performance metrics were: individual goals (IG), conceded goals (CG), goals by teammates (GT); and net goals (NG). Cognitive functions assessment aimed to measure cognitive flexibility (CF), impulsivity (I), sustained attention (SA), visuospatial working memory (VWM), and tracking capacity (TC). k-means clustering was applied to segment on field performance metrics into two groups: inferior and superior performance players. Cognitive functions were mixed in 31 datasets with all combinations for on-field performance prediction. Seven supervised machine-learning classification algorithms were tested. The standout algorithms were K-nearest neighbors and neural networks (Multilayer Perceptron). The best-supervised machine learning models presented balanced accuracies ranging from 69% to 72%. For predicting IG, CF, and TC stood out. SA and VWM presented better results in predicting CG. Regarding GT, SA, I, and CF provided better predictions. Lastly, the combination of CF, I, and VWM, precisely the executive functions, presented superior results in predicting NG, a measure reflecting athletes\' overall effectiveness. Cognitive functions demonstrated the sensitivity to differentiate athletes\' performance in small-sided games and highlighted the cognitive functions that presented better prediction for the different soccer game demands. (AU)

• FAPESP's process: 21/15134-9 - Can cognitive functions predict athlete's overall effectiveness and contribution to the team in small-sided soccer games? A Machine Learning approach
• Grantee: Rafael Luiz Martins Monteiro
• Support Opportunities: Scholarships in Brazil - Master