Abstract
The COVID-19 has adversely affected almost all countries of the world in a few months span and most of the countries are not able to manage the adverse impacts of it. There is a need for study that gives some help in management of this pandemic in future. Keeping this objective, this research evaluated the forecast performance of three ARIMA type models; Simple ARIMA, Sutte-ARIMA, and Wavelet-ARIMA for the COVID-19 data sets of Pakistan from 26 February 2020 to 13 July 2020. The data is separated into two portions: training and testing data. Training data ranges from 26th Feb, 2020 to 6th July, 2020 with 132 data points, for three series confirmed cases, deaths, and recoveries, respectively. On the other hand, testing data comprised of seven observations ranges from 7th July, 2020 to 13th July, 2020 for possible data validation process. The testing data is used to check the forecasting accuracy of the models used in this study. This study has compared three ARIMA type models predictive performances based on our data. Prediction performance for the next 7 days of testing data is applied through 3 different error measurements (MAE, MAPE, RMSE). Based on our findings, it is observed that Sutte-ARIMA performs best as compared to other two ARIMA type models, while Wavelet ARIMA performs worst. This study will help policy makers to make future management in better ways to keep away people from adverse effect of such pandemics.
