Dominance relation-based feature selection for interval-valued multi-label ordered information system

Multi-label learning addresses situations where a instance is linked to several labels. Existing multi-label feature selection has mainly addressed single-valued problems, while research on attribute reduction for interval-valued multi-label systems has yet to be reported. And explore how to apply the dominance principle to interval-valued multi-label ordered data is a promising area for future research. In this article, a new feature selection method was introduced, aiming to identify a more relevant and compact subset of features by incorporating label correlations and the dominance principle. First we combine multi-label learning with interval-valued information systems and design a new information system. Second, to make knowledge representation simpler, we discuss the dominance principle of interval-valued multi-label information systems. On this basis, we present a novel method for generating reduction information for each label and introduce a label correlation learning approach that exploits the overlap of this reduction information. Subsequently, an innovative feature selection algorithm utilizes dominance-based rough set is developed to efficiently filter out redundant features in the feature space. Finally, extensive experiments on nine multi-label datasets were performed, and the results confirm that the proposed algorithms surpass six state-of-the-art methods in performance and exhibit robustness.