采用进化计算的高维多模态特征选择

High-Dimensional Multimodal Feature Selection Based on Evolutionary Computation

针对特征选择问题中的多模态特性和高维特性,引入新的种群初始化策略对大规模多模态多目标优化算法进行改进,提出了一种基于进化多目标优化的高维多模态特征选择算法。对原始的连续优化算法进行离散化处理,用于评价离散优化问题中的个体,并在6个高维特征选择数据集上进行验证。结果表明:所提算法提升了初始种群的质量并加快了算法的收敛;相比于其他同类算法,所提算法获得了更优的帕累托前沿,其超体积指标值整体最优,并且在不影响分类精度的前提下可获得平均2.53个等效特征子集,表明所提算法具有最好的分类精度和最多样化的等效特征子集。

Aiming at multimodal and high-dimensional characteristics in feature selection problems,a new population initialization strategy is introduced to improve the large-scale multimodal multiobjective optimization algorithm,and a high-dimensional multimodal feature selection algorithm based on evolutionary multiobjective optimization is proposed.The original continuous optimization algorithm is discretized to evaluate individuals in discrete optimization problems.This approach is validated across six high-dimensional feature selection datasets.Results demonstrate that the proposed algorithm improves the quality of the initial population and accelerates algorithm convergence.Compared to other algorithms,the proposed algorithm yields the superior Pareto front with the best overall hypervolume value.It can obtain an average of 2.53 equivalent feature subsets without compromising classification accuracy.These results show the proposed algorithm’s ability to achieve optimal classification accuracy and the most diverse equivalent feature subsets.