An effective initialization for Fuzzy PSO with Greedy Forward Selection in feature selection

Feature selection is a critical step in machine learning, especially when dealing with high-dimensional datasets. Traditional methods often face challenges related to computational complexity and scalability. This paper introduces a hybrid feature selection approach, Fuzzy PSO with Greedy Forward Selection (FPGFS), that combines the global exploration capabilities of Fuzzy Logic-Based Particle Swarm Optimization (Fuzzy PSO) with the local refinement strengths of Greedy Forward Selection (GFS). The method leverages fuzzy entropy and fuzzy mutual information to initialize the PSO population, ensuring a focused search in the feature space. FPGFS balances exploration and exploitation through PSO's global search and GFS's local optimization. Experimental evaluations on 16 UCI and 21 ASU benchmark datasets demonstrate significant improvements in classification accuracy and feature subset reduction compared to state-of-the-art methods. Statistical analysis further validates the robustness and efficiency of FPGFS in handling high-dimensional data, offering a reliable solution for feature selection across various domains.