Real-time forecasts and risk assessment of novel coronavirus (COVID-19) cases: A data-driven analysis

Tanujit Chakraborty; Indrajit Ghosh

The rationale behind the choice of RT as a potential model to find the important casual variables out of 10 input variables for the CFR estimates is the simplicity, easy interpretability, and high accuracy of the RT algorithm. We apply an optimal RT model to the dataset consisting of 50 different country samples and try to find out potential casual variables from the set of available variables that are related to the case-fatality rates. RT is implemented using 'rpart' [31] package in R with "minsplit" equals to 10% of the data as a control parameter. We have used RMSE, co-efficient of multiple determination (R 2 ), and adjusted R 2 (AdjR 2 ) to evaluate the predictive performance of the tree model used in this study [17] . An optimal regression tree is built with 7 variables with 'minsplit' = 5 with equal costs for each variable. The estimates of the performance metrics for the fitted tree are as follows: RMSE = 0.013, R 2 = 0.896, and AdjR 2 = 0.769. A variable importance list from the RT is given in Figure 7 and the fitted tree is provided in Figure 8 .