Predicting Elite Sprint Performance: The Superiority of Biomechanical Variables in Machine Learning Models

This study investigates the impact of biomechanical and demographic variables on short-sprint performance in elite athletes. A biomechanical variable is a quantifiable, measurable parameter used to describe motion, forces, or physical characteristics and in this study, biomechanical variables relating to sprinting were analyzed. Data from 666 athletes tested at the Norwegian Olympic Training Center (1995–2018) were analyzed to predict 30-m sprint times using linear regression, random forest, and XGBoost models. Models were trained and evaluated both with and without biomechanical variables, including theoretical maximal force (F0), velocity (V0), power (Pmax), and force application metrics (RFmax, DRF). Predictive accuracy improved significantly when biomechanical variables were included, with mean squared error decreasing by factors of 931 (linear regression), 61 (random forest), and 8.5 (XGBoost), reflecting the critical role biomechanical variables play. Shapley value analysis indicated that Pmax, V0, and RFmax were the most influential predictors, while demographic and sport-specific factors dominated when biomechanical inputs were excluded. Although Sex, Age, and Body Mass may seem like important factors to an athlete's sprint time, they are outweighed in significance by biomechanical variables, indicating that superior biomechanics can outweigh differences in sex, age, or body mass. These findings highlight the critical role of mechanical sprint profiling in explaining sprint performance and suggest that individualized training programs should prioritize the optimization of horizontal power and velocity characteristics to maximize acceleration ability.