Artificial intelligence and forecasting of death by COVID-19 in Brazil: A comparative analysis of the algorithms Logistic Regression, Decision Tree, and Random Forest

ABSTRACT

This work makes use of artificial intelligence to contribute with empirical evidence that help predict death by COVID-19, enabling the improvement of health protocols used in health systems in Brazil and providing society with more tools to combat COVID-19. Data from January to September 2021 for Brazil are used in order to predict death by COVID-19 based on the clinical status of patients who used the Unified Health System in the studied period, in which three classification algorithms were tried: Logistic Regression (LR), Decision Tree (DT), and Random Forest (RF). The LR, DT, and RF models had a mean accuracy of 76%, 76%, and 77% in predicting death, respectively. In addition, it was possible to infer that when patients reach a point that require the use of ventilatory support and ICU, added to age, their chance of dying of COVID-19 is greater.

KEYWORDS
COVID-19; SARS-CoV-2; Logistic models; Artificial intelligence; Machine learning