Hyperparameter Tuning of Machine Learning Algorithms Using Response Surface Methodology: A Case Study of ANN, SVM, and DBN

Table 2

Best hyperparameter setting and the test performance of ANN.

Grid search					Confirmation runs			PI coverage
Fold	HN	TC	LR	MAE (validation)	Avg MAE (validation)	SD of MAE (validation)	95% PI half-width
1	7	604	0.005095	0.9513	1.0543	0.0605	0.1517	Yes
2	6	802	0.005095	1.0333	1.1461	0.0571	0.1432	Yes
3	9	208	0.005095	0.9637	1.0327	0.032	0.0803	Yes
4	9	852	0.005095	0.9743	1.0706	0.0458	0.1149	Yes
5	5	1000	0.005095	0.9621	1.0731	0.0188	0.0472	No
6	3	10	0.01009	0.9569	1.0265	0.0441	0.1106	Yes
7	9	654	0.01009	1.1191	1.1956	0.0431	0.1081	Yes
8	6	505	0.02008	1.0143	1.2014	0.0713	0.1788	No
9	7	951	0.005095	0.9894	1.0938	0.0538	0.1349	Yes
10	2	604	0.015085	1.0633	1.2126	0.0606	0.152	Yes

Bold values indicate the best set of hyperparameters from Grid Search that leads to the minimal MAE (validation).