Mayfly Optimization with Deep Belief Network-Based Automated COVID-19 Cough Classification Using Biological Audio Signals

Abstract

Abstract The outbreak of the COVID-19 pandemic has made widespread testing a need for controlling the disease. Numerous recent investigations have shown that many people with COVID-19 show no outward signs of illness. As a result, these patients are more likely to unwittingly spread the virus because they will not take a COVID-19 test. In order to get tested, patients will need to travel to a lab, putting others at risk of exposure. Recent studies have shown that people with COVID-19 who are asymptomatic have distinctive coughs and breathing patterns compared to the general population. This prompted the study of cough and breath sounds in COVID-19 patients as a means of differentiating them from those of non-COVID lung infection patients and the general population. In this article, we present a robust, efficient, and extensible method for identifying symptomatic patterns in biological audio signals. Cough digitized audio files are subjected to spectral analysis via a stationary wavelet transform (SWT). The proposed model employs ADASYN technique to handle the class imbalance problem. Also, features like Mel-frequency cepstral coefficients (MFCCs), log frame energies, zero crossing rate (ZCR), and kurtosis are extracted. For classification process, deep belief network (DBN) model is utilized. Finally, mayfly optimization (MFO) algorithm is exploited for optimal hyper-parameter tuning of the DBN model. The experimental validation of the proposed model takes place using open access dataset. Proposed method is compared with other methods in terms accuracy, specificity, sensitivity, F1-Score, precision and recall. The experimental outcomes demonstrated the betterment of the proposed model over other recent state of art approaches.