An Information-Based Elite-Guided Evolutionary Algorithm for Multi-Objective Feature Selection
Ziqian Wang, Shangce Gao, Zhenyu Lei, Masaaki Omura · IEEE/CAA Journal of Automatica Sinica · 2024
Dear Editor, This letter is concerned with the evolution strategy for addressing multi-objective feature selection problems in classification. Previous methods suffer from limitations such as being trapped in local optima and lacking stability. To overcome them, we propose a novel eliteguided mechanism based on information theory. Firstly, an elite solution is generated through a dimension reduction strategy and incorporated to the initialization population. Then, a symmetrical uncertainty-based mutation operator is developed to implement local search after the crossover operator. Finally, a special crowding distance is utilized to analyze duplicates in the environmental selection. The effectiveness and superiority of the proposed method are verified on 20 datasets, including high-dimensional ones.