Analysis of cognitive development in the prediction of technical, tatical and physical variables in Under-13 to Under-20 soccer players in small-sided games: Drone tracking validation and machine learning application.

Abstract

Soccer is a collective sport of complex and dynamic nature where performance is multifactorial, involvig maturational, technical, tactical, physical, and cognitive aspects. The application of new technologies has allowed advances in data acquisition and analysis. This project aims to analyze and characterize the influence of cognitive development in the prediction of technical, tactical, and physical variables in under-13, under-15, under-17, and under-20 soccer players during small-sided games, using machine learning and validating drone tracking for performance analysis. Two studies will be conducted for this purpose. Study 1 aims to validate the acquisition of positional data through drone-recorded videos. The data collection for this study has already been completed, involving 8 participants who performed a circuit with movements relevant and specific to sports. They were recorded simultaneously by a drone and a 12-camera Vicon system in a laboratory. Residual neural network algorithms will be retrained to detect and track participants in the drone videos. The positional data obtained will be compared using the root mean square error and the mean absolute error. 192 soccer players from the under-13, under-15, under-17, and under-20 categories will participate in the second study, with 48 players per category. Cognitive functions such as sustained attention, cognitive flexibility, impulsivity, visuospatial working memory, and tracking capacity will be evaluated. To assess on-field performance, a protocol involving multiple 3x3 small-sided games will be used. From the positional data obtained by tracking videos recorded by drones, technical, tactical, and physical variables will be extracted. These variables will be predicted by analyzing cognitive functions using 7 supervised machine learning classification algorithms. The accuracy of the models and differences in variables between categories will be compared using analysis of variance (ANOVA). The significance level will be set at p<0.05. (AU)