Spectral Features-Based Machine Learning Approach to Detect SARS-COV-2 Infection Using Cough Sound

In this paper, a spectral features based automated techniques for the classification of Severe Acute Respiratory Syndrome coronavirus using cough audio sound is presented. The proposed technique has following major stages: pre-processing, feature extraction, feature representation, and classifications. COUGHVID dataset is used for this study which comprises cough audio data of both Corona Virus disease 19 (COVID-19) positive and healthy subjects. Different audio features such as Mel Frequency Cepstral Coefficients and Zero Crossing Rate were extracted and represented using different methods. We found that frame-level labeling and feature representation is providing the best accuracy. The feature vector was taken as input to the classifier with 5-fold cross-validation. Support Vector Machine, Random Forest, K Nearest Neighbor, and Light Gradient Boosting Method model are tested in this paper to classify the COVID-19 positive and healthy subjects achieving an accuracy of 88.52%, 88.91%, 98.34%, and 81.95%, respectively.