CPSORCL: A Cooperative Particle Swarm Optimization Method with Random Contrastive Learning for Interactive Feature Selection

Complex diseases arise from the intricate interactions between multiple genetic factors and environmental influences. Genome-Wide Association Studies (GWAS) are essential for uncovering the genetic underpinnings of complex diseases. GWAS seeks to uncover genotype-phenotype relationships by analyzing the correlations between genetic variations and specific traits in large-scale datasets. Despite its substantial contributions, GWAS struggles to fully capture disease complexity when considering isolated features. Hence, the selection of interactive features has emerged as an innovative strategy to uncover the genetic mechanisms of diseases. This study presents a Cooperative Particle Swarm Optimization method with Random Contrastive Learning (CPSORCL) developed specifically for interactive feature selection. CPSORCL employs three key strategies: an adaptive random contrastive learning strategy, a feature-weight-guided flipping strategy, and a deep search strategy. The adaptive random contrastive learning strategy dynamically adjusts the topological structure according to population evolution, promoting both competition and cooperation among particles to maintain diversity. The feature-weight-guided flipping strategy adjusts feature flipping probabilities dynamically, striking a balance between global exploration and local refinement within the solution space. The deep search strategy accurately identifies relevant features within the candidate pool, determining the final set of interactive features. Experiments are conducted on simulated datasets and an age-related macular degeneration dataset to compare CPSORCL with seven widely used methods. The results demonstrate the potential of CPSORCL for interactive feature selection, highlighting its promise as an alternative to traditional methods. The source code for CPSORCL is publicly available at https://github.com/CDMBlab/CPSORCL.