Publication Date: 2023/06/15
Abstract: The COVID-19 pandemic has caused significant disruptions to global health, so-ciety, and the economy. Rapid and accurate detection of COVID-19 is crucial in minimizing community outbreaks and controlling the spread of the virus. This study proposes an audio-based digital testing method for COVID-19, eliminat-ing the need for patients to travel to testing laboratories. By analyzing cough noises using machine learning and deep learning techniques, the presence of COVID-19 can be detected and classified. The study evaluates multiple ma-chine learning models on the Coughvid dataset and assesses their performance in terms of accuracy. The results reveal that gradient boost achieves the highest accuracy of 88.82%, followed closely by Xgboost with an accuracy of 88.53%. Decision tree-based models, such as the Voting Classifier and Adaboost, also exhibit strong performance with accuracies above 88%. Logistic Regression, Deep Belief Network, MLP, Random Forest, and CNN demonstrate accuracies ranging from 87% to 88%. However, Linear Discriminant Analysis, PCA, Au-toencoder, and Na ̈ıve Bayes achieve comparatively lower accuracies, suggesting potential limitations in capturing the complexity of the dataset. The proposed audio-based digital testing method offers a promising approach to COVID-19 detection, providing a non-invasive and cost-effective solution for widespread testing and monitoring. The findings highlight the importance of leveraging machine learning techniques in healthcare and pave the way for further ad-vancements in audio-based COVID-19 detection methods.
Keywords: COVID-19, Cough Diagnosis, Deep Learning, Machine Learning, CNNs, Ensemble Methods, Voting Classifiers, Coughvid Dataset.
DOI: https://doi.org/10.5281/zenodo.8041580
PDF: https://ijirst.demo4.arinfotech.co/assets/upload/files/IJISRT23MAY1971.pdf
REFERENCES