Modelagem e Previsão do Número de Casos de Covid-19 Utilizando Aprendizagem de Máquina

Abstract

The main goal of this work is related to the fact that there is no official method for predicting the number of cases of Coronavirus (COVID-19) offered by the Brazilian federal government. Despite the current availability of technology and modeling methods, there is no official way to estimate the number of future cases of the disease in the country. This prevents greater preparation and availability of resources for better conditions of service to citizens and the fight against the virus from being implemented. This work proposes to apply a method of forecasting new cases of the disease, using past data. The modeling technique to be used is called Random Forest, a machine learning method capable of working with linear and non-linear classification and regression problems. In order to predict the future number of Covid-19 cases, we used the relative search volume of the top symptoms of the disease provided by Google Trends. We believe that greater direct searches for these symptoms by Google users would increase the number of Covid-19 cases in later days. However, the model output using Google Trends information did not result in a useful model for forecasting. By using the number of registered cases as a predictor variable, but lagged by 7 days, we obtained a very satisfactory result, in which the model was able to explain 95.9% of the variance of the response variable.