Multi-label Feature Selection Based on Fuzzy Neighborhood Mutual Discrimination Index

• Main Authors: Wang, Chenxi, Chen, E, Ren, Mengli, Yu, Xiehua, Lin, Yaojin, Li, Shaozi
• Format: Conference Proceeding
• Language: English
• Subjects: Classification algorithms; Feature extraction; Feature selection; Indexes; Information entropy; Information technology; Label enhancement; Multi-label learning; Rough sets; Training
• ISSN: 2474-3828
• DOI: 10.1109/ITME56794.2022.00130

Multi-label feature selection advances the performance of multi-label classification models by eliminating irrelevant and redundant features. Most multi-label feature selection algorithms assume that the training set containing logical labels, which means that labels are equally important for instances. However, in practical applications, there are different importances of labels. To solve the problem, we proposes a multi-label feature selection method based on fuzzy neighborhood mutual discriminant index. Firstly, logical labels are transformed into label distribution by label enhancement. Secondly, the information entropy is referenced to the label distribution learning, the label neighborhood similarity matrix is constructed to describe the ambiguity between labels. Finally, linking the fuzzy neighborhood discrimination index under the selected feature subset with label neighborhood similarity matrix is proposed, which is used for judging the distinguishing ability of the feature subset to label. Comprehensive experiments on 6 multi-label datasets show that the algorithm outperforms other algorithms in classification performance.