Explainable Machine Learning for Predicting Competitive Swimming Performance from Biomechanical, Physiological, and Training-Load Variables

Background: Competitive swimming performance is influenced by the interaction of biomechanical efficiency, physiological responses, anthropometric characteristics, and training exposure. Conventional analyses may overlook complex relationships among these domains.Aims: This study aimed to develop an explainable machine-learning model for predicting 100-m swimming performance and to identify the variables contributing most strongly to prediction.Methods: This observational study included 150 competitive swimmers (86 males and 64 females). Data comprised anthropometric variables, stroke rate, stroke length, stroke index, swimming velocity, intra-cycle velocity variability, start and turn time, resting, mean and peak heart rate, rating of perceived exertion, oxygen saturation, weekly swimming distance, training frequency, session duration, session RPE, and weekly training load. The 100-m performance time was used as the dependent outcome. Descriptive analysis, correlation analysis, and multivariable regression were combined with machine-learning prediction. Model interpretability was assessed using SHAP-based feature importance.Result: Mean 100-m performance time was 56.9 ± 4.3 s, with values ranging from 48.00 to 65.19 s. Swimming velocity demonstrated the strongest bivariate association with performance time (r = −0.71, p < 0.001), followed by stroke index (r = −0.62, p < 0.001) and stroke length (r = −0.48, p < 0.001). The multivariable model explained 68% of performance variance (R² = 0.68). The machine-learning model achieved a test-set R² of 0.73, RMSE of 2.32 s, and MAE of 1.79 s. SHAP analysis identified swimming velocity, stroke index, intra-cycle velocity variability, stroke length, and turn time as the leading predictors.Conclusion: Explainable machine learning can integrate multidimensional swimming data while identifying practically meaningful predictors of 100-m performance.