Forecasting of COVID-19 in Malaysia: Comparison of Models

COVID-19 first struck in December 2019 in Wuhan, China and became a global pandemic. This led to a global catastrophic impact on society, economy and politics, forcing the world to close borders and total lockdown as serious precautions against the infectious plague. Many studies were implemented, prospective and retrospectively, to control the vicious spread. The rising number of COVID-19 cases was alarming and forecasting or predicting the future number of cases became crucial. Thus, forecast modelling where the future number of cases would be predicted based on past and present time-series data, became the most-sort-after method. The country of Malaysia was not spared. Hence, this study aims to determine the best forecasting model in predicting the 30 days of COVID-19 number of cases fluctuations starting from the 1st implementation of Movement Control Order (MCO) in Malaysia on 18th March 2020 until the end of November of the same year. This study compared performances of the univariate modelling techniques and Box-Jenkins methodology to estimate and validate forecasting models, which is then used to forecast future number of COVID-19 cases, from the best model selected. From four univariate models: Average Percent Change (APC), Single Exponential Smoothing (SES), Double Exponential Smoothing (DES), and Holt's method models; and four ARIMA models using Box-Jenkins Methodology: ARIMA(0, 1, 3), ARIMA(1, 1, 2), ARIMA(0, 1, 2) and ARIMA(2,1,1), Holt's method model is the best model with the least error (RMSE= 243.59 and MAPE= 27.7787), followed by SES model (RMSE=243.648 and MAPE= 27.7795). The 30-day forecast from the best model revealed that the pandemic trend would substantially increase. As Holt's method is best suited for linear trends, thus made it susceptible to the random influence and the smoothing constants (alpha and beta) are best for fast response towards the number of COVID-19 cases that change accidentally.