Machine learning-based classification of Taekwondo Poomsae side kick performance using kinematic parameters and physical characteristics.

Abstract

To develop and validate machine learning (ML) models for classifying Taekwondo Poomsae side kick (SK) performance using kinematic parameters and physical function characteristics. Forty collegiate Taekwondo Poomsae athletes performed SKs with both legs. Two models were developed: a kinematic model incorporating SK and pelvic tilt angles at face and maximal heights, and a physical function model including range of motion measurements and Y-balance test scores. Performance quality was assessed by an expert evaluator using standardised criteria. Five ML algorithms were tested, and their performance was evaluated using area under the curve (AUC) analysis. Random forest classifiers demonstrated excellent performance in both models (kinematic model: AUC = 0.930, accuracy = 89.3%; physical function model: AUC = 0.930, accuracy = 89.3%). In the kinematic model, SK angle at maximal height emerged as the strongest predictor. For the physical function model, Y-balance test composite score showed the largest impact. These findings represent a substantial improvement over conventional subjective assessment methods by providing quantifiable, objective classification with high accuracy. ML algorithms can effectively classify Taekwondo SK performance using both kinematic and physical function parameters. SK angle at maximal height and dynamic balance emerged as the most important predictors in their respective models, providing quantitative criteria for performance assessment.