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

Tanujit Chakraborty; Indrajit Ghosh

The predicted values for the training COVID-19 cases datasets of the proposed hybrid model for five countries are further used for model adequacy checking and based on actual and predicted test outputs, we computed RMSE and MAE for all the datasets and reported them in Table 2 . The performances of the proposed hybrid ARIMA-WBF model are superior as compared to the individual models for Canada, France, and the UK, whereas, for India and South Korea, our results are competitive with ARIMA. It is often true that no model can be universally employed in all circumstances, and this is in relevance with "no free lunch theorem" [33] . Even if in a very few cases hybrid ARIMA-WBF model gave lower information criteria values (in terms of RMSE and MAE for training data), we still can opt for the hybrid model given the asymmetric risks involved as we believe that it is better to take decisions based on a hybrid model rather than depending on a single one at least for 8 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.