Multi-label-Specific Features Learning Algorithm Based on Label Importance and Fuzzy Rough Set

Abstract

Label-specific features learning is a prominent research hotspot in the field of multi-label learning, which aims to construct a classification model based on the distinctive features of each label rather than the whole features. Existing approaches regarding label-specific features usually assume that the importance of each label to an instance is equal. However, this popular strategy might be suboptimal as the importance of labels actually is different. In this paper, a multi-label-specific features learning algorithm based on label importance and fuzzy rough set is proposed. First, the importance of labels is measured based on the similarity of instances, which not only preserves the ranking of relevant and irrelevant labels, but also follows the principles of smoothness and normalization. Second, the correlation between labels is analyzed, and label-specific features of each label are extracted through a fuzzy rough set model. Experiments on several public available data sets demonstrate the effectiveness of the proposed algorithm.