COVID-19 Diagnosis from Blood Gas Using Multivariate Linear Regression

Abstract

With the impact of the COVID-19 outbreak, almost all scientists and nations began to show great interest in the subject for a long time. Studies in the field of outbreak, diagnosis and prevention are still ongoing. Issues such as methods developed to understand the spread mechanisms of the disease, prevention measures, vaccine and drug research are among the top priorities of the world agenda. The accuracy of the tests applied in the outbreak management has become extremely critical. In this study, it is aimed to obtain a function that finds the positive or negative COVID-19 test from the blood gas values of individuals by using Machine Learning methods to contribute to the outbreak management. Using the Multivariate Linear Regression (MLR) model, a linear function is obtained to represent the COVID-19 dataset taken from the Van province of Turkey. The data set obtained from Van Yüzüncü Yıl University Dursun Odabaş Medical Center consists of blood gas analysis samples (109 positive, 1146 negative) taken from individuals. It is thought that the linear function to be obtained by using these data will be an important method in determining the test results of individuals. Gradient Descent optimization methods are used to find the optimum values of the coefficients in the function to be obtained. In the study, the RMSProp optimization algorithm has a success rate of 58-91.23% in all measurement methods, and it is seen that it is much more successful than other optimization algorithms.