Determining an effective short term COVID-19 prediction model in ASEAN countries

Abstract

The challenge of accurately short-term forecasting demand is due to model selection and the nature of data trends. In this study, the prediction model was determined based on data patterns (trend data without seasonality) and the accuracy of prediction measurement. The cumulative number of COVID-19 affected people in some ASEAN countries had been collected from the Worldometers database. Three models [Holt’s method, Wright’s modified Holt’s method, and unreplicated linear functional relationship model (ULFR)] had been utilized to identify an efficient model for short-time prediction. Moreover, different smoothing parameters had been tested to find the best combination of the smoothing parameter. Nevertheless, using the day-to-day reported cumulative case data and 3-days and 7-days in advance forecasts of cumulative data. As there was no missing data, Holt’s method and Wright’s modified Holt’s method showed the same result. The text-only result corresponds to the consequences of the models discussed here, where the smoothing parameters (SP) were roughly estimated as a function of forecasting the number of affected people due to COVID-19. Additionally, the different combinations of SP showed diverse, accurate prediction results depending on data volume. Only 1-day forecasting illustrated the most efficient prediction days (1 day, 3 days, 7 days), which was validated by the Nash–Sutcliffe efficiency (NSE) model. The study also validated that ULFR was an efficient forecasting model for the efficient model identifying. Moreover, as a substitute for the traditional R-squared, the study applied NSE and R-squared (ULFR) for model selection. Finally, the result depicted that the prediction ability of ULFR was superior to Holt’s when it is compared to the actual data.