ABSTRACT
Uzbekistan as well as the rest of the world faces the third wave of COVID-19 and uses machine learning algorithms to predict the adverse outcome during the admission of new patients. We collected the dataset of 1145 patients admitted to the Republican Center for Emergency Medicine. We use different machine learning models to predict the COVID-19 severe course. This study uses feature selection procedures based on statistical tests and the elimination of linearly dependent features. The resulting multilayer perceptron yields a ROC AUC of 86.9% on the test set outperforming other machine learning algorithms and several competing works. The model relies on easily collected features without blood laboratory testing. It increases an availability of the reliable risk prediction to developing countries. © 2022, Springer Nature Switzerland AG.