Modelagem autorregressiva não linear do impacto do distanciamento social nos casos de Covid-19 no estado de São Paulo, Brasil

Abstract

The first cases of the SARS-CoV-2 coronavirus disease (COVID-19) were detected in China in December 2019, potentially causing severe acute respiratory syndrome (SARS). The lack of preventive treatment and initially the absence of vaccines forced authorities to adopt strict policies of isolation and social distancing, such as school closures and restrictions on the use of public spaces. These measures aimed to slow down or prevent the spread of the disease but had the potential to generate economic, political, and cultural impacts. Consequently, the full understanding of the impacts of these measures on the spread of the disease still requires investigation. Therefore, this study proposes modeling based on a nonlinear autoregressive neural network with exogenous inputs (NARX) that aims to relate social distancing to the number of new cases and deaths from COVID-19, based on real data from São Paulo. The model is created using data from cell phone antennas that indicate the movement of mobile devices, obtained from the Intelligent Monitoring System of São Paulo. Experiments were conducted to calibrate the model by varying the number of layers and neurons per layer, the training algorithm, and the delay in the time series to be considered in the autoregressive model. Subsequently, hypothetical scenarios of social distancing were defined to evaluate the influence of this variable on the number of new cases and deaths from the disease, with results that reinforce the importance of these containment measures. The best results were obtained from standardized data with MSE=1.009e-06 to predict new cases.