Artificial Neural Networks to access curve behavior of COVID-19 in Brazil: A learning experience based on other countries

Abstract

The COVID-19 is considered a pandemic due to global contamination. Brazil lacks precision in estimating the virus’s behavior because it has been in the early stages, underestimating notifications of confirmed cases. This study aimed to diagnose the curve behavior of the confirmed cases of COVID-19 in Brazil, based on infected rates, considering the total population and the contaminated population in other countries. For greater accuracy in estimating the Brazilian curve of infected, the Artificial Neural Network structure estimates the population with confirmed cases, combined by the arithmetic mean with SEIR and other estimation methods including ARIMA, SARIMA, trend Holt, and additive Winter. The results showed that, despite maintaining the adopted restriction policies, Brazil tends to face a crisis with a contagion curve below that registered by critical cases such as Spain and the United States, indicating the possibility of the occurrence of 1,000,000 confirmed cases on the 82nd day.