The Usage of Neural Network on Prediction of Covid-19 on Brazil

Resumo

The development of smart dynamic systems through computational models has been used extensively in

recent years to predict behaviors, gaining prominence in the field of medicine with regard to the diagnosis, identifi-
cation and spread of diseases. Usually associated with a probabilistic or classificatory model, these algorithms gain

prominence in image diagnostics, acting in the identification of tumors, and in the prediction of dynamic behavior,
such as virus propagation, due to their computational demand and easy data acquisition. Therefore, this work aims
to use techniques of artificial neural networks to analyze the propagation behavior of COVID-19 in Brazil. For the
architecture of the algorithm, a matrix calculation software implemented with neural network algorithms was used
and the prediction model was built by reading data from infected, recovered and deaths from the initial days, with
limited temporal sampling of the infection in the world. The neural network used follows the Feed-Foward model,
trained with backpropagation through the Levenberg-Marquardt equation and was adjusted to predict the number
of infected, recovered and deaths over the days using a pattern of behavior analysis based on the respective rates of
infection, recovery and deaths. The results obtained for the forecast were validated through the conference with the
data presented by the health departments for the days for which the network was configured to forecast, showing
considerable precision, which increases as new data are inserted, thus showing a practical model for recurrence
and analysis.