A High-Dimensional Feature Selection Method via Selection and Non-selection Operators and Local Search Mechanism in Particle Swarm Optimization

Recently, the problem of high-dimensional feature selection (FS) has become a current research focus in the field of evolutionary algorithms (EAs). However, most EA-based FS methods still face challenges in effectively combining relevance to remove redundant features, leading to low search efficiency and difficulty in finding high-quality feature subsets. To address this issue, this paper proposes a particle swarm optimization algorithm based on selection and non-selection operators and a local search mechanism, denoted as SNSLS-PSO. First, we design selection and non-selection operators based on Relief-F and roulette wheel selection to improve the quality of the selected feature subsets. Second, we introduce a local search mechanism based on an adaptive mutation operator, thereby avoiding local optima and enhance population diversity. In addition, we enhance the quality of particle selection by designing mutation probabilities for gene positions. The experimental results indicate that, compared to state-of-the-art FS methods, our proposed SNSLS-PSO can more efficiently select feature subsets and demonstrate superior classification performance on 14 high-dimensional datasets.