Kernelized Fuzzy Rough Sets-Based Three-Way Feature Selection

Xingchen Liu, Liuxin Wang, Linchao Pan, Can Gao

Abstract: Feature selection is the process of selecting important features from a dataset. The feature subset formed by important features represents the features of the entire dataset to reduce the complexity of subsequent computations. In recent years, feature selection methods based on rough set theory have been continuously developed, and the approximate quality of kernelized fuzzy rough sets is a better method for evaluating features. However, the heuristic greedy strategy adopted by traditional methods is difficult to guarantee the quality of feature subsets. Based on the idea of three-way decision, this paper proposes fuzzy dependency-based three-way feature selection method. We expand the three potential feature subsets through a differentiated approach and reduce the redundancy among them. Ensemble learning is performed on the three feature subsets to improve the classification performance. The experimental results show that compared with the traditional greedy feature selection method, the proposed feature selection method produces better classification performance, which demonstrates its effectiveness.