BACKGROUND: Swimming performance is a complex, multidimensional phenomenon shaped by the interplay of physiological, neuromuscular, and biomechanical factors. Nonetheless, the majority of prior research has employed simplified methodologies, constraining the comprehension of integrated performance determinants. This study aimed to find the multifaceted factors of 100 m swimming performance in both in-water and dry-land, as well as to establish a comprehensible machine learning framework for performance analysis in age-group swimmers.; METHODS: Thirty highly trainedage-group swimmers (16 males, 14 females; age 14.0 ± 1.0 years) underwent an extensive test battery comprising 11 dry-land and 2 in-water domains (99 variables). The performance outcomes consisted of official 100 m short-course times for butterfly, backstroke, breaststroke, and freestyle strokes. The normality of the data was validated (Shapiro-Wilk p > 0.05). Relationships were evaluated by correlation analysis and principal component analysis. Predictive modeling was conducted utilizingLeast Absolute Shrinkage and Selection Operator (LASSO), Elastic Net, and Random Forest with cross-validation.; RESULTS: Critical velocity showed as the most significant predictor across all strokes (r = *0.87 to *0.93, p < 0.001), followed by grip endurance peak force (r = *0.83, p < 0.001) and squat jump height (r = *0.75, p < 0.001). Elastic Net models accounted for 88.9-99.9% of performance variance (R² = 0.889-0.999; RMSE = 1.27-2.65 s), while LASSO exhibited a R² between 0.838 and 0.950. Random Forest analysis identified in-water variables (critical velocity, tethering force) and grip endurance as significant contributors. Notably, freestyle and backstroke performance was more influenced by dry-land neuromuscular factors, whereas breaststroke relied more heavily on in-water propulsion variables. Feature importance derived via SHapley Additive exPlanations (SHAP) confirmed these stroke-specific contributions, providing interpretable insights into model predictions. Domain-specific models demonstrated significant explanatory power for both dry-land (R² = 0.876-0.965) and in-water (R² = 0.660-0.892) predictors. CONCLUSIONS: Swimming performance is influenced by both multidimensional and stroke-specific parameters, with critical velocity serving as a universal predictor and neuromuscular elements playing a role across different strokes. Interpretable machine learning offers a comprehensive framework for identifying critical performance factors and facilitates personalized, data-informed training approaches for young swimmers
© Copyright 2026 XVth International Symposium on Biomechanics and Medicine in Swimming Proceedings. All rights reserved.