A hybrid dual-population cooperative immune algorithm for efficient high-dimensional feature selection

High-dimensional data analysis is fundamental to numerous domains, yet it is frequently impeded by the curse of dimensionality, which complicates pattern recognition and model generalization. However, existing high-dimensional feature selection methods are often limited by slow convergence, insufficient population diversity, and premature stagnation. Therefore, this study proposes a hybrid dual-population cooperative immune algorithm (DPCIA) for high-dimensional feature selection. Specifically, a Fisher score-based filtering strategy and adaptive position update mechanism based on EBPSO are adopted to enhance population initialization. In addition, a dual-population cooperative framework inspired by the population division strategy of the Self-adaptive ɛ-Constrained Multi-objective (Saɛ-CMO) algorithm is incorporated into the clonal selection algorithm (CSA) to achieve balanced exploration-exploitation. Moreover, population diversity is enhanced through lethal mutations, random perturbations, and an elite exchange mechanism between populations. During the classification validation phase, a -Nearest Neighbors (KNN) classifier is employed to evaluate the performance of the feature subsets. Experimental results on multiple high-dimensional datasets demonstrate that DPCIA achieves favorable feature compression and robust classification performance, with acceptable computational overhead for high-dimensional feature selection tasks. The implementation code of the proposed DPCIA algorithm and all experimental scripts are stored in an anonymous public repository for peer review. The anonymous link will be replaced with the official GitHub repository after acceptance for publication. Anonymous repository: https://anonymous.4open.science/r/DPCIA-4F53 .