Attribute Reduction Method Based on Improved Granular Ball Neighborhood Rough Set

• Published in: 2022 7th International Conference on Cloud Computing and Big Data Analytics (ICCCBDA) pp. 13 - 16
• Main Authors: Xia, Han, Chen, Zizhong, Wu, YanMin, Qi, JinLi
• Format: Conference Proceeding
• Language: English
• Published: IEEE 22.04.2022
• Subjects: attribute reduction; Benchmark testing; Big Data; Classification algorithms; Cloud computing; Costs; granular ball computing; granular ball neighborhood rough set; neighborhood rough set; Rough sets; Search methods
• DOI: 10.1109/ICCCBDA55098.2022.9778889

Feature reduction is an important aspect of big data analytics today, and neighborhood rough set is a classic attribute reduction method. The traditional neighborhood rough set finds out the radius suitable for problem solving by specifying the radius of the neighborhood or using a grid search method. There are a lot of overlapping calculations in the neighborhood calculation, and the calculation complexity is high. The granular ball neighbor-hood rough set can adaptively generate different neighborhood radius, and obtains better algorithm efficiency and learning effect than the traditional neighborhood rough set. However, due to the overlap problem in the process of granular ball generation, the granular ball neighborhood rough set still exists the overlap calculation of neighborhoods. For this reason, this paper solves the problem of granular ball overlap by detecting heterogeneous ball overlap and splitting, and thus designs an improved granular ball neighbor-hood rough set attribute reduction method. Experiments show that the improved granular ball neighborhood rough set has higher classification accuracy on the public benchmark dataset than the classic neighborhood rough set and the standard granular ball neighborhood rough set.