Enhanced Feature Selection for Imbalanced Microarray Cancer Gene Classification Using Chaotic Salp Swarm Algorithm

Abstract

The healthcare sector requires intelligent solutions to manage vast microarray data, but challenges like high dimensionality, data imbalance, and computational complexity persist. This study addresses these by handling imbalanced microarray cancer gene datasets using Synthetic Minority Oversampling Technique (SMOTE) and enhancing the Salp Swarm Algorithm (SSA) with sinusoidal chaotic map initialization for improved population and control parameter diversity. The enhanced SSA is combined with Chi-square and Mutual Information filter methods to select top-performing genes from Ovarian, Colon, and Leukemia datasets, followed by refinement based on minimal error. Key contributions include chaotic initialization for better exploration, SMOTE for balanced classification, and a novel minimal-error gene subset selection. Compared to state-of-the-art methods, our approach achieves competitive performance, with 100% accuracy and F1 score across datasets while reducing gene counts (e.g., 4 genes for Colon). This promises to enhance cancer diagnosis and treatment, enabling targeted therapies and personalized medicine for improved patient outcomes.