Improved logistic regression combined with recursive feature elimination for investigating key influencing factors of Tornado Kick Turn in Wushu Routines

Amidst the rising global interest in Wushu, identifying core factors in martial arts techniques is vital for offering scientific and unbiased training, enhancing practitioners' skills. This study reviewed martial arts training literature, pinpointing gaps and deficiencies in existing research methods. Concentrating on the tornado kick, it gathered data on physical attributes and movements from 50 elite athletes and 50 amateurs. To tackle limitations due to a limited sample size, the research employed an innovative approach merging k-fold cross-validation and recursive feature elimination to bolster a logistic regression model. Analysis of classification accuracies and SHAP values enabled the model to select and prioritize the most significant variables. The logistic regression model, enriched with five key features, achieved a remarkable 100% mean classification accuracy through 10-fold cross-validation. Notably, there was a clear difference in SHAP values related to initial jump angular velocity between elite and amateur athletes, underscoring its importance in the tornado kick's performance. This enhanced model provides a robust basis for examining other leaping techniques, introducing fresh analytical insights and advancing the global understanding and appreciation of martial arts.