Abstract
The COVID-19 pandemic has resulted in over 700 million confirmed cases and more than 7 million deaths worldwide, indicating the urgent need for rapid and accurate diagnostic tools. Early and precise detection of COVID-19 variants is critical for controlling outbreaks and guiding public health interventions. However, conventional diagnostic methods, such as PCR and antigen testing, have limitations in sensitivity, processing time, and adaptability with respect to emerging variants. Additionally, the absence of gene expression-based datasets in many studies restricts the ability to analyse host immune responses, leading to suboptimal detection and prediction of disease severity. To address these challenges, this study proposes the Optimal Explainable Artificial Intelligence for Gene Expression-based COVID-19 classification (OXAI-GECC), leveraging the Internet of Things (IoT) – San Francisco COVID-19 dataset for robust and interpretable COVID-19 variant classification. The methodology integrates data preprocessing, followed by Local Interpretable Model-Agnostic Explanations (LIME)-infused Black Widow with Particle Swarm Optimization (LIME-BW-PSO) for efficient feature extraction. The Shapley Additive Explanations (SHAP) algorithm is then employed for feature selection, ensuring that the most relevant biomarkers contribute to the classification process. Finally, a Convolutional Liquid Neural Network (CLNN) is used for classification, providing enhanced accuracy and adaptability to dynamic gene expression patterns. Experimental results demonstrate that the proposed OXAI-GECC framework achieves, for the data set considered, 100% accuracy, 100% precision, 100% recall, and 100% F1-score, outperforming existing methods in COVID-19 classification.