Analysis of disease severity and mortality prediction using machine learning during COVID-19

This paper focuses on how machine learning (ML) algorithms and applications have been used to analyze disease severity and mortality prediction in COVID-19 research. In the past, simpler statistical and epidemiological methods were more commonly used by researchers and officials to predict the course of the pandemic. However, in recent years, the limitations, high costs, and time required for medical tests have become significant challenges in stopping the spread of COVID-19. Some improved statistical methods have been used to tackle these challenges, but they have only partially solved the problems at a certain quality level. On the other hand, machine learning offers a wide range of smart methods, frameworks, and tools to deal with problems in the medical field. In this paper, using public and clinical data from patients, the severity and risk of death are studied through different machine learning algorithms, and the most important features in this area are identified. The main innovation of this paper is the comparative analysis of different models for diagnosis using statistical data. First, the COVID-19 dataset is preprocessed, and then several well-known models in disease classification are used, and their accuracy is compared. This study helps healthcare centers and hospitals prioritize the allocation of medical resources based on the severity of patients' conditions and predict their chances of survival. With data from over one million patients and the evaluation of >12 models, the Logistic Regression model generally shows the highest accuracy for both class 0 and class 1. In various situations, this model achieved the highest accuracy for class 0 (97 %) and for class 1 (80 %). Therefore, it can be concluded that the Logistic Regression model performs best in diagnosing both classes.