Abstract / Summary
Background: Five-a-side soccer for the blind (F5) is a Paralympic sport in which there are various performance variables, including anthropometric factors and physical fitness. However, despite this recognition, there is little scientific literature on performance analysis using algorithms to identify the main variables in this sport. Objective: The objectives of this study were (i) to identify key performance indicators in F5 players based on body composition and physical fitness variables and (ii) to rank anthropometric and physical performance variables using machine learning (ML) algorithms. Materials and Methods: A cross-sectional study was conducted to characterize the body composition and physical fitness levels of F5 players. A total of 20 F5 players participated. To address the high dimensionality and multicollinearity of the anthropometric variables while mitigating the risk of overfitting, penalized regression models (Elastic Net and Lasso algorithms) were implemented. Results: While unadjusted bivariate correlations suggested associations between body composition and performance, none survived False Discovery Rate (FDR) correction (pFDR ≥ 0.05). However, regularized multivariate models isolated exploratory candidate variables. Cross-validated performance was null (R2cv = 0.00%) for countermovement jump (CMJ) jump height, CMJ power, COD 5-0-5 and 20 m sprint, and moderate only for speed 0–5 m (R2cv = 59.42%; RMSEcv = 0.08 s). Conclusions: No statistically significant differences in anthropometric, body composition, or physical fitness profiles were detected between attackers and defenders (p > 0.05). Penalized ML algorithms identified a small set of exploratory candidate variables associated with specific physical tasks. Due to the small sample size (n = 20), these models should be considered exploratory and hypothesis-generating.