Table 1. Results of model hyperparameter tuning.

Model	Model hyperparametersa	Hyperparameter spaceb	Total combination	Best combination
ARIMAc	p, d, q	q=[0,1,2,3] d=[0,1,2,3] p=[0,1,2,3,4]	80	p=2 d=1 q=2
Double ESd	trend damped	trend=['add', 'mul'] damped=[True, False]	4	trend= 'add’ damped=True
MLPe	n_input n_nodes n_batch n_diff	n_input = [2-5] n_nodes = [100, 150] n_batch = [1, 150] n_diff = [0,1, 2,4]	64	n_input=4 n_nodes=100 n_batch=1 n_diff=1
LSTMf	n_input n_nodes n_batch n_diff	n_input=[2-5] n_nodes=[100, 150] n_batch=[1, 150] n_diff=[0,1, 2,4]	64	n_input=4 n_nodes=100 n_batch=1 n_diff=1
CNNg	n_input n_diff	n_input=[2-5] n_diff=[0,1, 2,4]	16	n_input=4 n_diff=1

^aHyperparameters are model-specific settings tuned to optimize performance.

^bValues in brackets represent the range or options tested using grid search.

^cARIMA: autoregressive integrated moving average.

^dES: exponential smoothing.

^eMLP: multilayer perceptron.

^fLSTM: long short-term memory.

^gCNN: convolutional neural network.