Hybrid deep learning framework for accurate classification of high dimensional genomic data

High-dimensional genomic datasets often contain redundant, noisy, and sparse features that make accurate classification challenging for conventional deep learning (DL) models. Existing approaches generally fail to maintain interpretability and stability when confronted with heterogeneous genomic structures. To address these limitations, this study proposes a hybrid TabNet-CNN framework that combines attention-driven feature selection with adaptive convolutional refinement. The attention mechanism in TabNet highlights the most relevant genomic attributes, while the convolutional layers enhance localized feature interactions for accurate decision boundaries. Experimental results on multiple genomic datasets demonstrate superior performance in accuracy, AUC, and interpretability compared to state-of-the-art models. The proposed framework holds promise for real-world applications such as biomarker identification, disease subtyping, and clinical decision-support systems in precision medicine.