Systematic hybrid feature selection using optimal filter classifier pairing and parallel evaluation.

There is extensive literature on feature selection algorithms, which are typically categorized into filter, wrapper, and embedded methods. Each category has its own merits and limitations. As a result, researchers have increasingly focused on developing hybrid methods that combine the strengths of these approaches. However, the selection of algorithms in hybrid methods is often made without adequate justification, which limits their applicability and overall effectiveness. We propose a Systematic Hybrid Feature Selection framework with Optimal Filter-Classifier Pairing and Parallel Evaluation (SHFS-OFCP-PE), designed to provide a filter-agnostic and classifier-independent hybrid feature selection strategy. The hybridization process emphasizes selecting features with minimal redundancy and maximum relevance to class labels. It also accounts for improvements in model performance when new features are added. Additionally, we present a parallel framework based on a master-slave architecture to enhance scalability. We conduct extensive experiments on twelve hybrid combinations and compare the results with those of traditional and state-of-the-art methods. The findings confirm the superiority of the proposed hybrid methods across multiple performance metrics and are further validated through a post hoc Analysis of Variance (ANOVA) test.