A Novel Dual-Crossed Deep Learning with an Adaptive Metaheuristic Approach and Improved AlexNet Based Covid-19 Variant Detection

The genomic analysis of COVID-19 is essential for understanding its evolution, detecting mutations, and tracking its epidemiological spread. This study proposes a bioinformatics-driven framework integrating advanced preprocessing, feature extraction, and classification techniques. The preprocessing phase applies N-gram tokenization, Quality Filtering, and Min-Max Scaling to refine genomic sequences. For feature extraction, the framework integrates a Dual-Crossed Improved AlexNet with a K-Means Clustering-based Self-Adaptive Tuna Swarm Optimization (SA-TSO) algorithm to enhance feature learning and representation. The Improved AlexNet incorporates Batch Normalization (BN) for stable training and Global Average Pooling (GAP) to reduce overfitting and computational complexity. For classification, an Improved BiRNN is employed to provide precise mutation identification and viral strain categorization, model long-range connections and highlight important genomic patterns. This integrated approach enhances the ability to monitor COVID-19 mutations, track epidemiological patterns, and support public health interventions by providing deeper insights into viral evolution and transmission dynamics. The experimental findings show that the suggested method achieves the maximum precision with 70% on the learning rate (98.9) and high precision with 80% on the learning rate (99.1).