Statistical Model for Predicting the COVID-19 Pandemic in South and Southeast Asian Regions

Abstract

The disease caused by severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) has been put in the list of public health emergency of international concern (PHEIC) according to the rapid spreading and expantion of the diseases around the world, later named the coronavirus disease 2019 (COVID-19). With in a year, more than 150 million people were identified as SARS-CoV-2 infected cases, and over 3.23 million deaths were reported from COVID-19 weekly update, 24st January, 2021, since the pandemic announcement from the World Health Organization (WHO). Nowadays, many regions world wide have been facing the crises due to the pandemic of COVID19, also South and Southeast Asian regions. This study aims to construct a model for predicting COVID-19 using natural cubic spline function with equi space knot. The data used in this study were obtained from publicly available databases form Johns Hopkins University coronavirus resource center updated daily and located at GitHub run by Microsoft. The results found that the model fits the data extremely well, defined as that maximizes the r-squared value in India, Pakistan, Bangladesh, Nepal, Sri-Lanka, Philippines, Malaysia, Thailand, Vietnam, and Australia were 0.997, 0.981, 0.992, 0.975, 0.995, 0.957, 0.973, 0.939, 0.989, 0.881, 0.943 and 0.610, respectively. Moreover, model provided mean r-squared 0.936. To access the performance of the natural cubic spline model, apart from using the r-squared values, we compared the RMSE for training and testing data set. We found that the RMSE between these models was not much different. This might be an evidence that the models were not over- fitting. Moreover, the model provides forecasts of daily changes, which signaled when action is needed. Moreover, this model is routinely applicable to all such regions in the world and can be extended to accommodate additional predictors such as environmental and demographic variables. Conclusion, this model is routinely applied to all such regions in the world and can be extended to accommodate additional predictors such as environmental and demographic variables.