Overview and cross-validation of COVID-19 forecasting univariate models

Atchadé, Mintodê Nicodème; Sokadjo, Yves Morel

Researchers have been working with different models to forecast COVID-19 cases. Many of their estimates are not accurate. This study aims to propose the best model to forecast COVID-19 cumulative cases using a machine learning technic. It is a work that focused on time series univariate models because there are too many debates about the quality of the pandemic data. To increase the likelihood of the findings, we avoided many variables modeling and proposed a robust process to forecast COVID-19 cumulative cases. It will help international institutions to take optimal decisions about the world economy and response to the pandemic. Consequently, we used the data titled ”Coronavirus Pandemic (COVID-19)” from ”Our World in Data” about cases from 22 January 2020 to 30 November 2020. We computed Error Trend Season (ETS), Exponential smoothing with multiplicative error-trend, and ARIMA on the training data sets. In addition, we calculated the Mean Absolute Percentage Error (MAPE) per model. Among those models, we notice that ETS (with additive error-trend and no season) has the smallest MAPE statistics compared to the others. The findings revealed that with the ETS model we need at least 100 days to have good forecasts with a MAPE threshold of 1%