Table 3.

Optimized hyperparameters of machine learning models by GA and PSO optimizers and the search range for the prediction of hydrogen yield.

Hyperparameter	Range	Unoptimized	GA	PSO
Random forest (RF)
n_estimators	10–500	100	187	10
max_depth	1–50	NaN *	13	41
min_samples_split	2–10	2	3	2
min_samples_leaf	1–10	1	1	1
extreme gradient boosting (XGB)
learning_rate	0.01–0.5	NaN	0.402014	0.268485
n_estimators	50–500	NaN	408	500
max_depth	3–10	NaN	9	5
min_child_weight	1–7	NaN	5	1
gamma	0–0.5	NaN	0.141899	0.267668
subsample	0.5–1	NaN	0.989653	0.715084
colsample_bytree	0.5–1	NaN	0.975411	0.500000
Decision tree (DT)
max_depth	1–50	NaN	49	50
min_samples_split	2–50	NaN	8	4
min_samples_leaf	1–50	NaN	7	1
Support vector machine (SVM)
C	0.1–1000	1	462.771600	74.618075
epsilon	0.01–1	0.1	0.424860	0.141974
gamma	0.1–1		0.035418	0.045675
Categorical boosting regressor (CatBoost)
learning_rate	0.01–0.5	0.03	0.252239	0.439795
depth	4–10	6	7	5
l2_leaf_reg	1–10	3	7.786251	9.244647
Artificial neural network (ANN)
learning_rate_init	0.0001–0.1	0.001	0.010889	0.035365
hidden_layer_sizes	5–100	100	81.428966	36.580111
activation_function	identity, logistic, tanh, relu	relu	tanh	logistic
Gaussian process regression (GPR)
sigma	0.0001–55	1	0.315774	0.050060
kernel_function	RBF *, Matern	RBF	RBF	RBF

* NaN: Not a Number; RBF: Radial Basis Function.