Multi-Label Attribute Evaluation Based On Fuzzy Rough Sets

In multi-label learning task, each sample may be assigned with one or more labels. Moreover multi-label classification tasks are often characterized by high-dimensional and inconsistent attributes. Fuzzy rough sets are an effective mathematic tool for dealing with inconsistency and attribute reduction. In this work, we discuss multi-label attribute reduction within the frame of fuzzy rough sets. We analyze the definitions of fuzzy lower approximation in multi-label classification and give several improvements of the traditional algorithms. Furthermore, the attribute dependency function is defined to evaluate condition attributes. A multi-label attribute reduction algorithm is constructed based on the dependency function. Numerical experiments show the effectiveness of the proposed technique.