SARS-COV-2 (COVID-19) detection application via machine learning: comparative analysis and performance evaluation

Abstract

Introduction

The worldwide COVID-19 pandemic is a serious problem for people around the world, especially for healthcare professionals. Currently, two COVID testing techniques are in vogue which includes RT-PCR and Rapid Antigen test. The RT-PCR is an expensive detection technique, and it takes approximately 24 h to test and process the sample in order to formulate the results. The detection of the virus at an earliest with higher accuracy and at optimal price are the prerequisites that need to be addressed by any detection technique proposed by the researchers. The cheaper and quicker method is known as Rapid Antigen test or rapid test. However, these tests are less reliable, with an accuracy rate in some cases as low as 50%. This requires the development of detection methods that target large populations and reduce the rate of detection errors by alleviating the burden on the health care system.

Objective

In this article, we have developed several machine learning models, namely (logistic regression, SVM, K-NN, and NB) utilizing various algorithms to predict COVID-19 based on the exhibited symptoms. The primary aim is to enable quick and accurate testing of individuals affected by the coronavirus. RESULTS: Through our work, we have achieved diverse accuracies (91.38%, 54.77%, 95.49%, and 91.57%) while utilizing different algorithms. Consequently, we have implemented our model with the linear SVM, which has an accuracy of 95.49%, an F-score of 92.08%, precision of 93.01%, and a recall of 91.16% for real-time detection.