COVID-19 MORTALITY PREDICTION USING MACHINE LEARNING METHODS

Abstract

The paper reports the use of machine learning methods for COVID-19 mortality prediction. An open dataset with large number of features and records was used for research. The goal of the research is to create the efficient model for mortality prediction which is based on large number of factors and enables the authorities to take actions to avoid mass spread of virus to and reduce the number of cases and deaths. Feature selection was conducted in order to remove potentially irrelevant input variables and improve performance of machine learning models. The classic machine learning models (both linear and non-linear), ensemble methods such as bagging, stacking and boosting, as well as neural networks, is used. Comparison of efficiency of ensemble methods and neural networks compared to classic ML methods such as linear regression, support vector machines, K nearest neighbors etc. is conducted. Ensemble methods and neural networks show much greater efficiency than classical ones. Feature selection does not significantly affect the prediction accuracy.

The scientific novelty of this paper is the large number of machine learning models trained on the large-scale dataset with significant number of features related to different factors that can potentially affect COVID-19 mortality, as well as further analysis of their efficiency. This will assist to select the most valuable features and to become a basis for creating a software designed for tracking the dynamics of the pandemic.

The practical significance of this paper is that present study can be useful for authorities and international organizations in prevention of COVID-19 mortality increase by taking proper preventive measures.