Predicting mortality outcomes in individual COVID-19 patients using machine learning algorithms

Abstract

In late 2019, the COVID-19 disease emerged, caused by the SARS-CoV-2 virus, and has since spread worldwide, becoming a global pandemic and resulting in almost seven million deaths to date. In addressing this global crisis, artificial intelligence has played a crucial role, particularly through the development of predictive models using machine learning algorithms, which have been successfully applied to solving a multitude of problems across multiple scientific fields. The purpose of this paper is to identify the model, or models, with the highest accuracy in predicting a COVID-19 patient’s mortality outcome by comparing their performance metrics. Different ML methods employed in model development include logistic regression, decision trees, random forest, eXtreme gradient boosting (XGBoost), multi-layer perceptrons, and the k-nearest neighbors. The metrics used for the comparison of these models were accuracy, precision-recall, F1 score, area under the receiver operating characteristic curve (AUC-ROC), and runtime. The data used comprised the clinical characteristics and histories of 12,425,179 individuals who attended health facilities in Mexico. Following a comprehensive evaluation, the XGBoost model achieved the highest overall score across all metrics. It scored 93.76% in precision, 95.47% in recall, 91.13% in F1-score, 97.86% in AUC-ROC, and had a runtime of 6.67306 s. Therefore, XGBoost was determined to be the preferred method for predicting the mortality outcome of COVID-19 patients.