{"title":"A novel cooperative co-evolutionary algorithm with context vector enhancement strategy for feature selection on high-dimensional classification","authors":"Zhaoyang Zhang, Jianwu Xue","doi":"10.1016/j.cor.2025.107009","DOIUrl":null,"url":null,"abstract":"<div><div>Feature selection (FS) in high-dimensional classification is challenging due to the exponential increase in the number of possible feature subsets and the increased risk of model overfitting. Focusing on this challenge, the cooperative co-evolutionary algorithm (CCEA), which adopts a divide-and-conquer strategy, has been successfully applied to FS. However, existing CCEA based FS methods risk getting trapped in pseudo-optimum. To address this issue, we propose a novel CCEA based FS method, which can manipulate the context vector to escape pseudo-optimum. Specifically, competitive swarm optimizer (CSO), a variant of particle swarm optimization (PSO), is extended into CCEA based CSO due to its high performance in high-dimensional problems. This forms the foundation of the proposed method. Subsequently, a non-parametric space division strategy is proposed, which enables the proposed method to adaptively handle both low-dimensional and high-dimensional data. Most importantly, a context vector enhancement strategy is proposed, which performs search across all subspaces of the context vector, enabling proposed method to escape pseudo-optimum. Following this, a subpopulation enhancement strategy is proposed, which generates new individuals according to the search results of context vector enhancement strategy to replace individuals with poor fitness, accelerating the evolution of subpopulations. We compare the proposed method with a few state-of-the-art FS methods on 18 datasets with up to 12600 features. Experimental results show that, in most cases, the proposed method is able to find a smaller feature subset with higher classification accuracy.</div></div>","PeriodicalId":10542,"journal":{"name":"Computers & Operations Research","volume":"178 ","pages":"Article 107009"},"PeriodicalIF":4.1000,"publicationDate":"2025-02-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Computers & Operations Research","FirstCategoryId":"5","ListUrlMain":"https://www.sciencedirect.com/science/article/pii/S0305054825000371","RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q2","JCRName":"COMPUTER SCIENCE, INTERDISCIPLINARY APPLICATIONS","Score":null,"Total":0}
引用次数: 0
Abstract
Feature selection (FS) in high-dimensional classification is challenging due to the exponential increase in the number of possible feature subsets and the increased risk of model overfitting. Focusing on this challenge, the cooperative co-evolutionary algorithm (CCEA), which adopts a divide-and-conquer strategy, has been successfully applied to FS. However, existing CCEA based FS methods risk getting trapped in pseudo-optimum. To address this issue, we propose a novel CCEA based FS method, which can manipulate the context vector to escape pseudo-optimum. Specifically, competitive swarm optimizer (CSO), a variant of particle swarm optimization (PSO), is extended into CCEA based CSO due to its high performance in high-dimensional problems. This forms the foundation of the proposed method. Subsequently, a non-parametric space division strategy is proposed, which enables the proposed method to adaptively handle both low-dimensional and high-dimensional data. Most importantly, a context vector enhancement strategy is proposed, which performs search across all subspaces of the context vector, enabling proposed method to escape pseudo-optimum. Following this, a subpopulation enhancement strategy is proposed, which generates new individuals according to the search results of context vector enhancement strategy to replace individuals with poor fitness, accelerating the evolution of subpopulations. We compare the proposed method with a few state-of-the-art FS methods on 18 datasets with up to 12600 features. Experimental results show that, in most cases, the proposed method is able to find a smaller feature subset with higher classification accuracy.
期刊介绍:
Operations research and computers meet in a large number of scientific fields, many of which are of vital current concern to our troubled society. These include, among others, ecology, transportation, safety, reliability, urban planning, economics, inventory control, investment strategy and logistics (including reverse logistics). Computers & Operations Research provides an international forum for the application of computers and operations research techniques to problems in these and related fields.