MOPSO-ES: Multi-objective particle swarm optimization algorithm based on niche technology and dynamic elliptical segmentation with distance screening mechanism for high-dimensional feature selection

Feature selection (FS) is a core preprocessing technique for improving data quality, and it often needs to strike a balance between the conflicting goals of maximizing classification accuracy and minimizing feature dimension. A multi-objective particle swarm optimization algorithm (MOPSO-ES) based on niche technology and elliptical segmentation dynamic distance screening (ES) mechanism is proposed. Firstly, the DBSCAN clustering algorithm is introduced to adaptively partition the population; Then, combined with niche technology and the ES mechanism, the traditional grid partitioning strategy is replaced, effectively enhancing the algorithm's coverage ability in the Pareto frontier boundary area; Finally, a dynamic distance selection strategy is applied during the non-dominated solution maintenance stage. This strategy analyzes the domain characteristics of continuous or binary problems to select the corresponding distance formula for filtering and removing solutions, thereby enhancing population diversity. In the experimental section, the ES mechanism is firstly applied to five grid-based multi-objective algorithms (MOGWO, PESA-II, MOWOA, MODA and MOBA). Its effectiveness is verified by using six standard multi-objective test functions, and ablation experiments evaluate the independent contribution and synergistic value of each module. Next, MOPSO-ES is applied to 12 high-dimensional datasets to solve feature selection problems. It is compared with nine other multi-objective algorithms (NSGA-II, NSGA-III, WOF, VaEA, MOEA/D, MOCOA, EDEFS, LR-GPSOFA and MOHHOFOA). The results show that MOPSO-ES achieves competitive performance on most high-dimensional datasets. In addition, a sensitivity analysis of key hyperparameters is provided and the performance stability of the proposed method under different classifiers is tested.