as it is known to be an adaptive learning rate optimization algorithm designed specifically for training deep neural networks [32]. The learning rate was set initially at 0.005 and was decreased by a factor of 0.2 at every 200 epochs. The number of maximum epochs was chosen to be 1000. In order to avoid exploding gradients effect, a threshold 1 was set as the gradient threshold. ... the decomposed IMFs with similar center frequencies are used to train separate RNNs. Here we further work out the phase of each decomposed IMF corresponding to the CC, T, and H signals for both states using Gabor's complex analytical signal. V. Results and Discussion Figure 14 shows the predicted number of cases of COVID-19 for CC-IMF1 (Figure 14(b)), CC-IMF2 (Figure 14(c)) and CCIMF3 (Figure 14(d)) and the sum of all of them (Figure 14(a)) for the state of Maharashtra conducted on the test set. The figures show the Root Mean Square Error (RMSE) corresponding to each case. The results show that the model can predict the future number of cases within an acceptable range of error. However, Figure 14(b) shows that the predicted value of the signal deviates from its expected value at the right end of the signal. This is likely due to the end effect arising from the spline method used in the smoothing procedure to smooth the feature TR. This effect is also evident in Figure 2(c). As can be seen in the figure, the right end of the signal is slightly tilted downward whereas this is not the case in the original signal of Figure 2(a). Therefore, one may argue that smoothing the 120 180 160 100 140 120 Phase (rad) Phase (rad) 80 60 40 20 -20 0 10 20 30 40 50 Days (a) 60 70 80 60 0 -20 90 90 160 80 140 70 120 Phase (rad) 50 40 30 20 CC-IMF2 T-IMF2 H-IMF2 10 0 -10 0 10 20 30 40 50 Days (c) 60 70 80 CC-IMF3 T-IMF3 H-IMF3 20 60 Phase (rad) 80 40 CC-IMF2 T-IMF2 H-IMF2 0 100 0 10 20 30 40 50 Days (b) 60 80 90 100 80 60 40 CC-IMF3 T-IMF3 H-IMF3 20 0 90 70 -20 0 10 20 30 40 50 Days (d) 60 70 80 90 FIGURE 16 Unwraped phase corresponding to the IMFs of CC, T and H signals of states Maharashtra and Tamil Nadu. (a) Maharashtra (IMF2). (b) Maharashtra (IMF3). (c) Tamil Nadu (IMF2). (d) Tamil Nadu (IMF3). NOVEMBER 2020 | IEEE COMPUTATIONAL INTELLIGENCE MAGAZINE 47