Attribute reduction for multi-label classification based on labels of positive region

摘要

In this paper, on the basis of the rough set theory, four attribute reduction algorithms are proposed for multi-label data. In order to improve the computational efficiency, the proposed algorithms utilize the lower approximations of the label information set instead of the decision class to evaluate the importance of attributes. The relationship between the proposed methods and two classical attribute reductions is analyzed and shows that the proposed methods are more applicable to multi-label classification. Experimental results reveal that the proposed algorithms can remove redundant attributes without reducing classification accuracy for most data.