Title
Modeling COVID-19 Confirmed Cases Using a Hybrid Model
Authors
Abstract
The COVID-19 virus has caused numerous problems worldwide. Given the negative effects of COVID-19, this study aims to estimate accurate forecasts of the number of confirmed cases to help policymakers determine and make the right decisions. This paper uses a hybrid approach for forecasting the daily COVID-19 cases based on combining the Autoregressive Integrated Moving Average (ARIMA) and Autoregressive Neural Network (NNAR) with a single hidden layer. To fit the linear pattern from the data, ARIMA models are used. Then, the NNAR models are used to capture the nonlinear pattern. The final prediction is obtained by adding up the two predictions. Using six-time series from January 22, 2020, to June 22, 2021, of new daily confirmed cases of COVID-19 from Pakistan, Tunisia, Indonesia, Malaysia, India and South Korea, this work evaluates the hybrid approach against some benchmark models and generated ten days ahead forecasts. Experiments demonstrate the superiority of the hybrid model over the benchmark models. Given the complex nature of new confirmed cases, it is assumed that the data contains both linear and nonlinear components. In literature, different studies have tended to forecast future cases of COVID-19. However, most of them have used single models that capture either linear or nonlinear patterns. This paper proposes a hybrid model that captures both linear and nonlinear components from the data.
Keywords
COVID-19, Time series forecasting, Autoregressive Integrated Moving Average, Autoregressive Feedforward Neural Network.
Classification-JEL
Pages
128-162