Prediction of Covid-19 contagion in the State of Maranhão using Neural Networks and the SIR epidemiological model

Resumo

Coronavirus disease 2019 (COVID-19) is a public health emergency of international interest, being
declared by the World Health Organization (WHO) as a pandemic, and has instituted essential measures for
prevention and confrontation. Because it has high rates of transmission and rapid dissemination around the world,
medical authorities worldwide have realized its potential danger and the possible collapse of the Health Systems

of the affected countries. Therefore, specific predictive methods are urgently needed to predict the risk of COVID-
19 and to assist in the decision-making process in the control of epidemics and / or pandemics. Here, the number

of cases for the State of Maranhão in the period from 27/03 to 27/05 of this year was estimated, with observations
up to the Lockdown decree using the epidemiological model SIR (Susceptible-Infected-Recovered) and a neural
network Multilayer Perceptron (MLP). The samples were obtained through Kaggle, collected from the Ministry
of Health. In order to compare the algorithms, the population was divided into three groups: susceptible, infected
and recovered. MLP averaged 5507.95 cases, while the actual number was 5787.56. An epidemiological
mathematical model estimated the number of cases with a larger lag (about 1.8 million cases). The results obtained
by the proposed method can be used to support COVID-19 decision making. The neural network has shown to
have more reliability in the results and is closer to the real cases disclosed until May 17, 2020.