Abstract
This study aimed to determine the contribution of classic anthropometric features in young soccer players to predict their speed of preplanned change of direction performance using explainable artificial intelligence methodologies. Fifty-one young soccer players (age: 11.88 ± 1.38 years, body mass: 38.25 ± 7.89 kg) from a professional youth soccer academy were recruited. Several anthropometric features, including calf girth, leg length, sitting height, thigh length, and skinfold thicknesses, were employed following standardized protocols. Players' change of direction speed was assessed using the 505-test. Biological maturity was estimated using the Mirwald equation. Five machine learning algorithms were implemented, with decision tree regression (DTR) selected as the optimal approach. Model validation employed leave-one-out cross-validation. The 505-test demonstrated excellent reliability (ICC = 0.86, SEM = 3.54%). Using the DTR model, anthropometric features accurately predicted change-of-direction speed performance (P < .05). The DTR model achieved superior predictive accuracy (R2 = 0.843, RMSE = 0.133, MAE = 0.111), explaining 84.3% of the variance in performance. DTR analysis revealed that demographic and anthropometric features, including sitting height and leg length, significantly affected change of direction speed performance (overall Gini importance > 0.5). Key predictors included maturity offset (16.3%), age (16.8%), leg length (15.8%), sitting height (13.6%), and knee height (10.5%). Youths who are more mature and have shorter sitting heights achieved superior change-of-direction speed performance. These findings underscore the importance of considering anthropometric and maturity characteristics in male youth soccer players to support talent identification, providing evidence-based frameworks for development programs.