A Comparative Study of Machine Learning Methods to Predict COVID-19

Abstract

First appearing in Wuhan City, Hubei region, China, the COVID-19 disease has threatened public health, trade, and the global economy. The World Health Organization has recommended testing for COVID-19 using a Reverse Transcription Polymerase Chain Reaction (RT-PCR) protocol to address diverse viral genes. Nevertheless, these test protocols demand RNA extraction kits, expensive machines, and trained technicians to operate them. Therefore, alternatives that are faster to diagnose, cheaper, and easier to access for patients and medical personnel are needed. This chapter presents a comparative analysis of machine-learning techniques for detecting COVID-19. The following four classifiers were trained, tested, and compared using the cross-validation technique with five folds: Random Forest, Stochastic Gradient Descent, Naive Bayes, and K- Nearest Neighbors. The dataset used in this project was the one the Government of Mexico has made available on the Internet on the Datos Abiertos Dirección General de Epidemiología web page. The results indicate that the Random Forest classifier performs best based on the area under the curve and the precision-recall curve metrics.