Forecasting of Covid-19 Using Time Series Regression Models

Abstract

The novel coronavirus (COVID-19) pandemic is a major global health threat that is spreading very fast around the world. In the current study, we present a new forecasting model to estimate the number of confirmed cases of COVID-19 in the next two weeks based on the previously confirmed cases recorded for 62 countries around the world. The cumulative cases of these countries represents about 96% of the total global up to the date of data gathering. Seven regression models have been used for two rounds of predictions based on the data collected between February 21, 2020 and August 31, 2020. We selected five feature sets using various feature-engineering methods to convert time series problem into a supervised learning problem and then build regression models. The performance of the models was evaluated using Root Mean Squared Log Error (RMSLE). The findings show a good performance and reduce the error about 70%. In particular, XGB and LGBM models have demonstrated their efficiency over other models.