Inter-session consistency of sprint-derived mechanical outputs in youth male soccer players: A combined classical and data-driven approach

This study examined the between-session reliability of sprint-derived horizontal force–velocity (FV) profile variables in forty highly trained youth male soccer players (age 13.63 ± 1.03 years), and explored whether machine-learning (Random-Forest) analyses complement traditional reliability statistics. The participants completed 25-m linear sprint tests across two sessions, using five-split-times to derive FV variables. Reliability was assessed using intraclass correlation coefficients (ICC (3,1) ), concordance correlation coefficients (CCC), standard error of measurement (SEM), smallest worthwhile change (SWC), and minimal detectable change (MDC 95 ). AI-based analyses were performed using Random Forest regression to examine model-based consistency through the prediction of retest values from test data, and Random Forest classification to assess multivariate session separability. Results differed across FV profile variables. Velocity- and force-effectiveness measures, including theoretical maximal velocity (V 0 ), ratio of force (RF), and decrease in ratio of force, demonstrated greater stability than force- and power-related variables, such as theoretical maximal force (F 0 ), horizontal force, theoretical maximal power (P 0 ), and power. The FV slope (SFP) differed significantly between sessions and was therefore excluded from subsequent analyses despite showing good relative reliability (ICC = 0.82, 95% CI: 0.66–0.90; CCC = 0.79, 95% CI: 0.56–0.91), with SEM = 0.06, MDC95 = 0.17, and MDC95% = 15.10%. Random-Forest-regression showed strong predictive accuracy for several FV variables (R 2 ≥ 0.80), consistent with ICC and CCC results, whereas performance was weaker for power and short sprint variables (R 2 = 0.37). FV profile reliability was variable-dependent, and Random Forest machine learning analyses supported the findings obtained from classical reliability statistics.