Cough2COVID-19 detection using an enhanced multi layer ensemble deep learning framework and CoughFeatureRanker

In response to the pressing requirement for precise and easily accessible COVID-19 detection methods, we present the Cough2COVID-19 framework, which is cost-effective, non-intrusive, and widely accessible. The conventional diagnostic methods, notably the PCR test, are encumbered by limitations such as cost and invasiveness. Consequently, the exploration of alternative solutions has gained momentum. Our innovative approach employs a multi-layer ensemble deep learning (MLEDL) framework that capitalizes on cough audio signals to achieve heightened efficiency in COVID-19 detection. This study introduces the Cough2COVID-19 framework, effectively addressing these challenges through AI-driven analysis. Additionally, this study proposed the CoughFeatureRanker algorithm, which delves into the robustness of pivotal features embedded within cough audios. The CoughFeatureRanker algorithm selects the most prominent features based on their optimal discriminatory performance from 15 features to detect COVID-19. The effectiveness of the CoughFeatureRanker algorithm within the ensemble framework is scrutinized, confirming its favorable influence on the accuracy of COVID-19 detection. The Cough2COVID-19 (MLEDL) framework achieves remarkable outcomes in COVID-19 detection through cough audio signals, boasting a specificity of 98%, sensitivity of 97%, accuracy of 98%, and an AUC score of 0.981. Our framework asserts its supremacy in precise non-invasive screening through an exhaustive comparison with cutting-edge methodologies. This groundbreaking innovation holds the potential to enhance urban resilience by transforming disease diagnosis, offering a significant approach to curtailing transmission risks and facilitating timely interventions in the ongoing battle against the pandemic.