Precision Diagnosis in Pandemics: Machine Learning and Cough Sound Analysis for COVID-19

The unprecedented global crisis caused by the COVID-19 pandemic has significantly impacted public health, economies, and daily routines. A critical challenge in managing the pandemic is promptly and precisely identifying COVID-19 cases. As a result, there is an urgent need for alternative, non-invasive, and swift diagnostic techniques to identify infected individuals and mitigate the spread of the virus. In this study, we introduce a machine-learning method for early COVID-19 diagnosis by analyzing cough sounds. The study utilizes Mel-frequency cepstral coefficients, extracted from cough audio. Extensive experiments using the COUGHVID cough dataset demonstrate the feasibility and effectiveness of the approach which uses the Extra-Tree classifier. The classification model outperforms other models with a testing accuracy of 90.05%, sensitivity of 86.01%, and an area under the curve of 90.08%, outperforming other similar works, in terms of performance and computation.