An interactive filter-wrapper multi-objective evolutionary algorithm for feature selection

作     者：Liu, Zhengyi Chang, Bo Cheng, Fan 

作者机构：Anhui Univ Sch Comp Sci & Technol Key Lab Intelligent Comp & Signal ProcMinist Edu Informat Mat & Intelligent Sensing Lab Anhui Prov Hefei 230601 Peoples R China 

出 版 物：《SWARM AND EVOLUTIONARY COMPUTATION》 (群与进化计算)

年 卷 期：2021年第65卷

页      面：100925-100925页

核心收录：

学科分类：08[工学] 0812[工学-计算机科学与技术（可授工学、理学学位）] 

基　　金：Natural Science Foundation of Anhui Province [1908085MF182] Key Program of Natural Science Project of Educational Commission of Anhui Province [KJ2019A0034] Humanities and Social Sciences Project of Chinese Ministry of Education [18YJC870004] University Synergy Innovation Program of Anhui Province [GXXT-2020-050] National Natural Science Foundation of China 

主　　题：Feature selection Multi-objective evolutionary algorithm Guiding strategy Repairing strategy Initialization strategy 

摘      要：As an important task in data mining, feature selection can improve the performance of classification by eliminating the redundant or irrelevant features in original data. It is mainly divided into filter method and wrapper method, and each one has its own advantages. To make full use of the advantages of two methods, in this paper, an interactive filter-wrapper multi-objective evolutionary algorithm, named GR-MOEA is proposed, where guiding and repairing strategies are used to select feature subsets with high quality. To be specific, a wrapper population and a filter population are evolved simultaneously in the proposed algorithm. To utilize the merits of two populations, an interactive scheme is designed, which includes a wrapper to filter guiding strategy and a filter to wrapper repairing strategy. The guide strategy is to use the good solutions in the wrapper population to steer the filter population towards a better direction. While in the repairing strategy, some features in the wrapper population are repaired by the useful information in filter population, which can avoid the trapping of local optimum in wrapper population. To further enhance the performance of GR-MOEA, two effective initialization strategies are also developed. Empirical studies are conducted by comparing the proposed algorithm with several state-of-the-art on different datasets, and the experimental results demonstrate the superiority of GR-MOEA over the comparison methods in obtaining the feature subsets with higher qualities.