Multi-label feature selection based on label distribution and neighborhood rough set

Jinghua Liu, Yaojin Lin, Weiping Ding, Hongbo Zhang, Cheng Wang, Jixiang Du

Abstract: Highlights•We extend the NRS model that can be skillfully applied to figure out label distribution data and avoid the instability of results caused by different neighborhood granularity selection.•Based on the nearest neighbor information distribution, a label enhancement method is designed to transform logical labels into label distribution, thereby obtaining more potential label information to guide feature selection.•A novel forward algorithm of multi-label feature selection based on the extended neighborhood rough set is proposed, which leverages feature significance, label significance, and label-specific features, simultaneously.