In recent years, the expansive utilization of data mining techniques has revolutionized various fields, including the analysis of dance videos. This study leverages data mining to meticulously capture and analyze dance movements, thereby facilitating the enhancement and correction of dancers’ techniques. Within the scope of this research, images of dance movements extracted from videos are subjected to preprocessing, which involves grayscaling and thresholding, to prepare them for further analysis. Building on these processed images, this paper introduces a novel multi-feature dance action recognition approach. This method integrates several distinct features—directional gradient histogram features, optical flow directional histogram features, and audio features—employing a linear weighting scheme within a multi-core learning framework. The efficacy of the proposed approach is demonstrated through its performance on the FolkDance dance dataset, where it achieves a 3.5% increase in fusion accuracy over the traditional Dance Style Identification (DSI) method. Additionally, when compared with the Multi-Feature Learning-Combined (MFL-C) method, our approach shows an improvement of 0.6% in fusion accuracy. This research establishes a viable method for the recognition and classification of dance movements, laying a robust foundation for further inquiry and practical applications in this domain.