Multi-Label Feature Selection Based on Min-Relevance Label

Multi-label feature selection has been widely adopted to address multi-label data with high-dimension features. It is critical calculate label correlations for selection. Existing methods adopt different schemes correlations, which obtain importance of labels. However, there exist two issues regarding these calculating importance: first, previous cannot predict the whole labels well because they only focus on most important labels; second, have similar classification information corresponding redundant To this end, we use mutual metric cores set rather than Afterwards, capture features respect each core label, finally, obtaining an optimal subset. verify effectiveness, our method compared state-of-the-art 16 real-world sets several evaluating metrics. The results experiment proves that proposed achieves best performance among all methods.