Thank you for the comment. The sample code was not rigorously optimized. I have fine-tuned the epochs and the number of LSTM layers. Now the actual and predicted values in the plots are in line with each other.