Biomechanical and Body Composition Factors in Shot Put Performance: A Predictive Model Using Machine Learning

Abstract

This study examines the key biomechanical and body composition features influencing shot put performance, utilizing machine learning models to predict shot distances. Four models Random Forest, Gradient Boosting, Categorical Boosting, and extreme Gradient Boosting were employed to analyze a dataset of 42 elite athletes. Fifteen biomechanical features were assessed for importance using the Random Forest model. Through feature selection, release velocity, gender, shot path length, and body mass emerged as the four most influential predictors of shot put performance, while shot release height, technique, and angle of release were among the least influential factors. Model performance was evaluated using Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), and the coefficient of determination (R²). Of the models tested, Gradient Boosting showed the highest predictive accuracy, achieving an R² of 0.8248, an MAE of 0.4474, and an RMSE of 0.6500. Following hyper parameter tuning, the final model was evaluated on unseen data, demonstrating impressive predictive accuracy and further validating its robustness. These findings provide valuable insights into the relationship between biomechanical and body composition factors and shot put performance, offering practical applications for athletes and coaches seeking data-driven approaches to optimize performance. By utilizing the model developed in this study, athletes and coaches can use their own data to predict shot distance, enabling more targeted and effective training strategies.