Revisiting Modified Auto-encoder for Prediction of the Dynamics of Covid-19
Machine Learning, Artificial Neural Network, Auto-encoder, Pandemic, COVID-19
Due to the declaration of the worldwide pandemic caused by the spread of the SARS-COV-2 virus, also called COVID-19, governments, institutions and researchers around the world mobilized to try to mitigate the effects caused by the effects of the virus in society. Several types of approaches have been proposed and employed in an attempt to predict the behavior of indicators that have some kind of connection with the pandemic. Among these methodologies, the models that use data orientation are known as data-driven models, in which they obtained considerable prominence among the others. Artificial Neural Networks are a type of model significantly disseminated within data-driven models. In this work, a new architecture of an ANN called Auto-Encoder is proposed. This new architecture aims to make time series predictions related to the COVID-19 pandemic, in particular the number of deaths. For this, other time series are used that may be directly related to what you want to predict. As inputs, time series corresponding to the number of cases, temperature, humidity and air quality (Air Quality Index - AQI) for the city of São Paulo, Brazil, were used. The partial results obtained demonstrate that the proposal has a promising accuracy in predicting the time series regarding the number of deaths in COVID-19.