Novel Framework for Robust Gene Selection and Accurate Multi-Cancer Classification

Abstract

This study presents the ensemble adaptive gene selection and classification framework (EAGSCF), a novel method for cancer classification using high-dimensional gene expression data. EAGSCF integrates hybrid feature selection, adaptive dimensionality reduction, and ensemble deep learning to address challenges such as high dimensionality, class imbalance, and interpretability. By combining mutual information (MI), recursive feature elimination, and the least absolute shrinkage and selection operator (LASSO), the framework extracts a compact, biologically meaningful subset of features. Meanwhile, uniform manifold approximation projection and variation auto encoders (VAEs) enhance their capacity to capture nonlinear relationships, which are crucial for distinguishing complex cancer subtypes. With top accuracy across four cancer datasets—98.9% for lung, 98.5% for colon, 98.2% for prostate, and 97.8% for lymphoma—EAGSCF outperforms existing methods, demonstrating significant potential in biomarker discovery and clinical use.