{"title":"SMEM: A Subspace Merging Based Evolutionary Method for High-Dimensional Feature Selection","authors":"Kaixuan Li;Shibo Jiang;Rui Zhang;Jianfeng Qiu;Lei Zhang;Lixia Yang;Fan Cheng","doi":"10.1109/TETCI.2024.3451695","DOIUrl":null,"url":null,"abstract":"In the past decade, evolutionary algorithms (EAs) have shown their promising performance in solving the problem of feature selection. Despite that, it is still quite challenging to design the EAs for high-dimensional feature selection (HDFS), since the increasing number of features causes the search space of EAs grows exponentially, which is known as the “curse of dimensionality”. To tackle the issue, in this paper, a <bold>S</b>ubspace <bold>M</b>erging based <bold>E</b>volutionary <bold>M</b>ethod, termed SMEM is suggested. In SMEM, to avoid directly optimizing the large search space of HDFS, the original feature space of HDFS is firstly divided into several independent low-dimensional subspaces. In each subspace, a subpopulation is evolved to obtain the latent good feature subsets quickly. Then, to avoid some features being missed, these low-dimensional subspaces merge in pairs, and the further search is carried on the merged subspaces. During the evolving of each merged subspace, the good feature subsets obtained from previous subspace pair are fully utilized. The above subspace merging procedure repeats, and the performance of SMEM is improved gradually, until in the end, all the subspaces are merged into one final space. At that time, the final space is also the original feature space in HDFS, which ensures all the features in the data is considered. Experimental results on different high-dimensional datasets demonstrate the effectiveness and the efficiency of the proposed SMEM, when compared with the state-of-the-arts.","PeriodicalId":13135,"journal":{"name":"IEEE Transactions on Emerging Topics in Computational Intelligence","volume":"9 2","pages":"1712-1727"},"PeriodicalIF":5.3000,"publicationDate":"2024-09-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"IEEE Transactions on Emerging Topics in Computational Intelligence","FirstCategoryId":"94","ListUrlMain":"https://ieeexplore.ieee.org/document/10665901/","RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q1","JCRName":"COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE","Score":null,"Total":0}
引用次数: 0
Abstract
In the past decade, evolutionary algorithms (EAs) have shown their promising performance in solving the problem of feature selection. Despite that, it is still quite challenging to design the EAs for high-dimensional feature selection (HDFS), since the increasing number of features causes the search space of EAs grows exponentially, which is known as the “curse of dimensionality”. To tackle the issue, in this paper, a Subspace Merging based Evolutionary Method, termed SMEM is suggested. In SMEM, to avoid directly optimizing the large search space of HDFS, the original feature space of HDFS is firstly divided into several independent low-dimensional subspaces. In each subspace, a subpopulation is evolved to obtain the latent good feature subsets quickly. Then, to avoid some features being missed, these low-dimensional subspaces merge in pairs, and the further search is carried on the merged subspaces. During the evolving of each merged subspace, the good feature subsets obtained from previous subspace pair are fully utilized. The above subspace merging procedure repeats, and the performance of SMEM is improved gradually, until in the end, all the subspaces are merged into one final space. At that time, the final space is also the original feature space in HDFS, which ensures all the features in the data is considered. Experimental results on different high-dimensional datasets demonstrate the effectiveness and the efficiency of the proposed SMEM, when compared with the state-of-the-arts.
期刊介绍:
The IEEE Transactions on Emerging Topics in Computational Intelligence (TETCI) publishes original articles on emerging aspects of computational intelligence, including theory, applications, and surveys.
TETCI is an electronics only publication. TETCI publishes six issues per year.
Authors are encouraged to submit manuscripts in any emerging topic in computational intelligence, especially nature-inspired computing topics not covered by other IEEE Computational Intelligence Society journals. A few such illustrative examples are glial cell networks, computational neuroscience, Brain Computer Interface, ambient intelligence, non-fuzzy computing with words, artificial life, cultural learning, artificial endocrine networks, social reasoning, artificial hormone networks, computational intelligence for the IoT and Smart-X technologies.