Tamil Nadu (Figures 16(c) and 16(d)). This suggests a more complex dependency among IMFs of these signals corresponding to Tamil Nadu compared with Maharashtra. One way of achieving more accuracy in prediction in the case of Tamil Nadu is to use a deeper RNN architecture. However, in order to avoid over-fitting, either more data or a more severe regularisation strategy has to be exploited. A further example is now investigated, corresponding to the state Gujarat in India where the number of cases is smaller. The data from this state is of interest particularly due to an outlier presenting at around day 54 (Figure 17(a)). The transmission rate TR of Figure 17(b) has been smoothed using the technique proposed in Section II. Also, all the signals CC, T, and H are decomposed using VMD (Section III-B), and their stationary and non-stationary parts are grouped and used for training the RNNs as discussed respectively in Sections III-B and IV. Figure 18 shows the final results of the prediction process. A satisfactory value of MAPE = 4.68% is obtained, which further confirms the applicability of the proposed technique. VI. Conclusion A systematic procedure to derive features for training RNNs to forecast the future number of confirmed cases of COVID-19 Forecast Cases Cases 500 450 0 2 4 6 10 8 12 14 16 RMSE = 30.4753 2 4 0 10 8 12 14 16 18 12 14 16 18 12 14 16 18 14 16 18 RMSE = 11.8983 10 5 0 2 4 6 10 8 12 14 16 0 18 0 2 4 6 10 8 Month (a) Month Forecast Forecast (b) 50 Cases 20 0 -20 0 2 4 6 10 8 12 14 16 0 -50 18 0 2 4 6 10 8 RMSE = 25.1369 RMSE = 13.8487 40 50 20 Error Error 6 15 50 Error Error 0 20 40 Cases 500 450 18 100 -50 Forecast 550 550 0 0 -50 -20 0 2 4 6 10 8 12 14 16 18 0 2 4 6 10 8 Month Month (c) (d) Observed 12 Forecast FIGURE 18 Predicted value and Root mean square error (RMSE) of the number of COVID-19 cases forecast corresponding to each IMF of Gujarat CC signal conducted on the test set (MAPE = 4.68%). (a) CC; (b) CC-IMF1; (c) CC-IMF2; and (d) CC-IMF3. NOVEMBER 2020 | IEEE COMPUTATIONAL INTELLIGENCE MAGAZINE 49