Attribute Reduction with Fuzzy Weighted Neighborhood Rough Sets

Abstract

Neighborhood rough sets (NRSs) have been widely recognized as an efficient tool for solving attribute reduction problems in numerical data. In constructing neighborhood information granules, many existing extensions introduce weighting mechanisms for conditional attributes to quantify their influence on the decision attribute, while other studies focus on assigning weights to individual objects within granules. Nevertheless, the internal distribution of objects inside each granule, particularly the density of objects in the same decision class, has not been sufficiently characterized. Therefore, the generated information granules may not accurately assess the actual contribution of individual objects. To address this issue, a generalized weighted neighborhood rough set model has recently been proposed by integrating both object weights and attribute weights. In this study, we investigate a special case of that framework by proposing an alternative approach to assigning weights to objects that better captures local class-consistency and structural characteristics within neighborhood granules. Based on the newly constructed information granules, we develop the Fuzzy Weighted Neighborhood Rough Set (FWNRS) model as a basis for designing an efficient attribute reduction method. Experimental evaluations on multiple benchmark datasets indicate that the proposed method is effective and outperforms several state-of-the-art approaches in terms of reduct size as well as classification performance.