Forecast of COVID-19 New Cases Using an MLP Regressor

Abstract

Over the past year the COVID-19 pandemic has overwhelmed healthcare systems and government institutions worldwide, creating the need for accurate prediction of confirmed cases. Current research focuses on prediction methodologies aiming at mitigating economical anxiety and aiding in detection, preemption and forecasting of the pandemic. Modeling using artificial neural networks (ANN) is an important component of this effort and this research aims to contribute with a methodology to forecast occurrence of new cases of COVID-19 1, 2 and 7 days ahead by using historical case data. These approaches were compared and contrasted with some published results in terms of network architecture and neural network of use. Three regressors were developed and showcased in this document with their accuracy evaluated using the root mean squared error (RMSE) from 1179 to 2806. The quality evaluation supports the conclusion that these regressors compete with those of current architectures.