Classification of COVID-19 Cough Sounds using Mel Frequency Cepstral Coefficient (MFCC) Feature Extraction and Support Vector Machine

Abstract

A lot of research has been carried out to detect COVID-19, such as swabs, rapid antigens, and using x-ray images. However, this method has the disadvantage that it requires taking samples through physical contact with the patient. One way to avoid physical contact is to use audio through coughing with the aim of reducing the transmission of COVID-19. Audio feature extraction such as the Mel Frequency Cepstral Coefficient (MFCC) has often been used in audio classification research, such as the classification of musical genres and so on. This study aims to compare more or less the features of audio classification performance through coughing sounds for early detection of COVID-19 using a Support Vector Machine based on the Linear and Radial Basis Function (RBF). The dataset used is the COVID-19 Cough audio dataset, before classifying, the audio data is processed into a spectrogram and then feature extraction is carried out. Classification is divided into 2 schemes, using default parameters, then using the specified configuration parameters. From the research results, the highest AUC is 0.572266 in the linear kernel-based SVM classification. Meanwhile, when using the RBF kernel, the highest AUC is 0.560181.