Explainable quality assessment of effective aligned skeletal representations for martial arts movements by multi-machine learning decisions.

How to utilize modern technological means to provide both accurate scoring and objective feedback for martial arts movements has become an issue that needs to be addressed in the field of physical education. This study proposes an intelligent scoring method based on machine learning. Firstly, the key features are extracted by the feature alignment technique, which eliminates the influence of athletes' movement speed, rhythm and duration on the scoring, thus reflecting the athletes' skill level more realistically. Second, to further improve the objectivity and accuracy, an adaptive weighted multi-model decision-making strategy is proposed. In addition, this study is the first to use interpretable artificial intelligence to provide feedback for teaching and learning Wushu. Experimental results indicate that the integrated model using the weighted average strategy not only outperforms other algorithms after feature alignment (on the XSQ dataset MAE is 0.237, RMSE is 0.442, sMAPE is 8.569, R[Formula: see text] is 0.633, Pearson's correlation is 0.807, ICC is 0.856. on the UMONS-Taichi dataset, MAE is 0.29, RMSE is 0.438, sMAPE is 10.01, R[Formula: see text] is 0.844, Pearson correlation is 0.921, and ICC is 0.954. on the PBB dataset, MAE is 0.261, RMSE is 0.351, sMAPE is 6.765, R[Formula: see text] is 0.557, Pearson correlation is 0.753, and ICC is 0.82), but is also close to the performance of human experts. In conclusion, this study not only achieves a performance of movement evaluation comparable to human experts, but also provides a technical framework for the rapid realizing of automatic scoring in other martial arts styles, which will promote the popularization and development of martial arts education.