Fig. 2.
Results of hyperparameter optimization with fivefold cross-validation. A grid search method was used for hyperparameter optimization. A total of 18 combinations have been tested for the number of memory cells in long short-term memory and the number of nodes in the hidden layer of feed-forward neural network. The combination representing the smallest validation error was selected as the hyperparameter of the model (8 memory cells and 16 nodes).