PSO-UFS: A Novel Approach to Univariate Feature Selection Using Particle Swarm Optimization

Abstract

Abstract Univariate Feature Selection (UFS) traditionally involves a labor-intensive process of trial-and-error, necessitating the selection of scoring functions and the determination of feature numbers. These choices can inadvertently affect both the performance and interpretability of the model. To address this challenge, we introduce Particle Swarm Optimization for Univariate Feature Selection (PSO-UFS), an innovative method that automates these crucial decisions. PSO-UFS leverages the power of Particle Swarm Optimization (PSO) to autonomously identify the optimal scoring function and feature subset that maximize a machine learning algorithm's performance metric. Our empirical evaluations across multiple datasets demonstrate that PSO-UFS significantly outperforms traditional UFS in various performance metrics, including accuracy, precision, recall, and F1-score. Importantly, PSO-UFS generates more interpretable feature subsets, thereby enhancing the model's comprehensibility. This advancement paves the way for broader applications in real-world scenarios where feature reduction and interpretability are paramount.