IEWGBNRS: A novel attribute reduction approach based on information entropy-weighted granular-ball neighborhood rough set

The granular ball neighborhood rough set model is an improved rough set model that integrates granular computing and neighborhood relations, capable of effectively handling continuous-valued data. However, existing models struggle to reasonably evaluate attribute importance under inconsistent attribute weights, leading to the risk of mistakenly removing key attributes during attribute reduction. To address this issue, we propose an attribute reduction algorithm for the granular ball neighborhood rough set model based on information entropy weighting. First, attribute weights are assigned by combining the internal distribution characteristics of attributes with the correlation between conditional and decision attributes. Second, an improved model based on the weighted granular ball neighborhood relation is constructed, and the corresponding attribute reduction method is defined. Finally, attribute reduction is implemented based on the average purity of granular balls. Extensive experiments are conducted on multiple UCI datasets. The results show that the proposed method not only achieves a high attribute reduction rate (up to 76.92%), but also significantly improves the average classification accuracy of classifiers such as Support Vector Machine (SVM) and K-Nearest Neighbors (KNN), increasing by 1.57% and 1.58%, respectively. Moreover, the method demonstrates stronger adaptability and stability in processing high-dimensional and complex datasets, indicating promising application potential.