采用进化计算的高维多模态特征选择

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.