Modelagem do número de novos casos confirmados por dia da COVID-19 no Brasil com uso de LSTM e predição linear

• Karhyne P. Assis UFABC
• Camila M. Silva UFABC
• Kenji N. Filho UFABC
• Ricardo Suyama UFABC
• André K. Takahata UFABC

DOI: https://doi.org/10.5753/ercas.2021.17429

Abstract

We analyzed the behavior of unit step predictors to predict the number of reported cases of COVID-19 per day. We investigated predictors created with the use of long short-term memory (LSTM) neural networks and we assessed their performance in comparison to linear predictors. We identified cases in which LSTM performs better, but also some challenges to make the LSTM based predictors capable of generalizing its performance.

References

Dong, E., Du, H. and Gardner L. (2020) “An interactive web-based dashboard to track COVID-19 in real time”, In: Lancet, 20(5), 533-534.

Oliveira, W. K., Duarte, E., França, GVA. And Garcia, LP. (2020) “How Brazil can hold back COVID-19”, Epidemiol Serv Saude; 29(2): e2020044.

Romano, J. M. T., Attux, R. R. F., Cavalcante, C.C., Suyama, R. (2011) “Unsupervised signal processing: channel equalization and source separation”. CRC Press.

Sabino, E. C., et al. (2021) “Resurgence of COVID-19 in Manaus, Brazil, despite high seroprevalence”, In: Lancet, 397(10273), 452-455.

Shertinsky, A. (2020) “Fundamentals of recurrent neural network (RNN) and long shortterm memory (LSTM) network”, Physica D: Nonlinear Phenomena, v. 404, p. 132306.