Estudo de algoritmos de redes neurais recorrentes para predição de casos e óbitos diários pela Covid-19 no Centro-Oeste

Abstract

Because of losts and sequels caused by the dacases and daily deaths as a result of COVID-19 disease placed on Midwest Region in Brazil, the predictive models of Recurrent Neural Networks, the Long Short Term Memory and Gated Recurrent Unit, were evaluated from studies obtained in related works, using as database the informations provided by the Coronavírus Brasil (2021) site from Ministry of Health. Using pre-processing techniques to optimize the dataset and using the hyperparameters presented in related works as a basis, it was possible to obtain good results for both models. The Long Short Term Memory model performed better than the Gated Recurrent Unit model, in relation to daily cases, resulting in the following performance metrics: Mean Absolut Error 963.92; Root Mean Squared Error 1261.53 and r2_score 0.94. On the other hand, the Gated Recurrent Unit model got better results for daily deaths, relying on the results in the performance metrics: Mean Absolut Error 29.07; Root Mean Squared Error 40.10 and r2_score 0.96.