Guided Particle Adaptation PSO for Feature Selection on High-dimensional Classification

The Particle Swarm Optimization (PSO) algorithm, renowned for its efficiency and ease of implementation, is widely utilized in solving NP-hard problems, including feature selection. However, in high-dimensional data scenarios, most existing PSO-based feature selection methods typically employ a single filter-based approach for initializing particle populations, limiting the search range. We propose a guided particle adaptation method that integrates several filter-based methods to create guiding particles. These particles play a beneficial guiding role and expand the search range. Moreover, we introduce a new fitness factor promoting knowledge transfer under particle guidance, preventing premature convergence to global optima. Results demonstrate that this method efficiently obtains high-precision feature subsets.