Preliminary diagnosis of COVID-19 based on cough sounds using machine learning algorithms

Abstract

The proposed work is focused on COVID-19 classification of cough sounds based on machine learning which is used to differentiate COVID-19 coughs from non COVID-19 and healthy coughs. It follows a non-contact based screening test which is very easy to apply being non-invasive and simply carried out within the boundaries of home so that the medical testing centers are not over flooded with patients and there is an overwhelming pressure because of maintenance of those patients with shortage of adequate infrastructure facilities. The dataset used in this study has been derived from the Coswara database which comprises of around 160 infected and 480 healthy individuals. Therefore, Artificial Intelligence based machine learning classifiers were used as an alternative means of diagnosis. Logistic regression (LR), K- Nearest neighbor (KNN), support vector machines (SVM), decision tree algorithms were used as classifiers in the proposed work. The results of this study show that the SVM classifier turned out to be the best in comparing among the COVID-19 and non COVID- 19 coughs with area under receiver operating characteristic curve (ROC) of 0.98. The novelty in the proposed work includes the collection of dry cough samples which would aid in preliminary diagnosis of the infection. This form of classification can also be implemented in a smart phone after performance evaluation from medical authorities.