A Case Study on Forecasting New Daily Cases of COVID 19 at Different Scales in Brazil


Forecasting new COVID-19 daily cases is a task that favors actions that mitigate the damages caused by the virus. In this perspective, the present work evaluates different Machine Learning models for this purpose, considering different scales in Brazil (country, state, and city levels). After a grid search with 860 configurations, the results indicate an Artificial Neural Network well suited for national level forecasting. On the other hand, with a lower performance at the larger scale, Elastic Net was good at predicting smaller ones. This case study highlights the difficulties of model reusing for COVID-19 forecasting and also the necessity of model choice and adjustment depending on the data scale.

In: XXXIX Simpósio Brasileiro de Telecomunicações e Processamento de Sinais