High-resolution three-dimensional biomechanical analysis and machine learning–based prediction of stroke length–stroke rate interaction in competitive freestyle swimming

Abstract

Background and Study Aim. Competitive freestyle swimming performance is determined by the coordinated interaction of multiple biomechanical variables. Although stroke length and stroke rate are widely used to evaluate swimming technique and performance, their interaction in relation to swimming velocity remains a subject of practical interest. This study aimed to investigate the relationship between swimming velocity, stroke length, and stroke rate using high-resolution three-dimensional biomechanical analysis and to develop an interpretable predictive model. Materials and Methods. A cross-sectional biomechanical study was conducted involving 40 competitive freestyle swimmers. Three-dimensional kinematic data were obtained using a Vicon motion capture system operating at 200 Hz. Regression modelling techniques, together with the XGBoost machine learning algorithm, were used to examine the interaction between stroke length and stroke rate and to predict swimming velocity. Results. The interaction between stroke length and stroke rate was a significant predictor of swimming velocity (β = 0.42, SE = 0.08, 95% CI: 0.26–0.58, p < 0.001). The nonlinear regression model explained a substantial proportion of the variance in swimming velocity (R² = 0.71; adjusted R² = 0.67). Conclusions. A nonlinear interaction between stroke length and stroke rate contributes to variations in competitive freestyle swimming performance. These findings suggest that optimizing the coordination between these biomechanical variables, rather than maximizing either variable independently, may enhance swimming performance.