Cough Sound based COVID-19 Detection with Stacked Ensemble Model

Abstract

COVID-19 pandeamic has affected people all over the world. COVID-19 may manifest with different severity in different people, however, it predominantly affects respiratory system. Symptoms may vary from sore throat and cough to shortness of breath and damaged lungs. This work focusses on developing a smart system for early detection of COVID-19 based on cough sounds and machine learning algorithms. Such a system would be easily accessible and may provide initial screening for detection of COVID-19. Moreover, cough sounds may be recorded by the person on smartphone avoiding the need for visiting a hospital or testing facility and getting exposed to the disease during the pandeamic. First, the duration of cough sound is determined in the recorded audio signal using thresholding. Then, statistical features are extracted for cough sound and normalized. Finally, the performance of 10 different machine learning algorithms are compared for automatic detection of COVID-19. The proposed stacked ensemble of machine learning models yields the best performance, with an accuracy of 79.86% and area under region of convergence curve of 0.797 for cough sounds of new patients.