关键词: 特征选择, 多标记分类, 聚类融合, 互信息

Abstract: In the framework of multi-label learning, feature selection is a powerful tool for solving the curse of dimensionality, which can improve the classification performance of multi-label classifier. In this paper, a multi-label feature selection algorithm via fusing  feature ranking is proposed. First, it?conducts adaptive graining samples based on different labels and employs the neighborhood of sample to compute the neighborhood mutual?information between feature and label, which can measure the importance degree of feature. Then, all features are?sorted in?descending order by the value of their neighborhood mutual?information under each label. Finally, it acquires a new feature rank by fusing all individual feature rank lists. Experiment is conducted on four data sets, and four evaluation?criteria are used to measure the effectiveness. Experimental results show that the proposed algorithm is superior to several state-of-the-art multi-label feature selection algorithms.

Key words: feature?selection, multi-label classification, clustering ensemble, mutual?information