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

Table 9

Best hyperparameter setting and the validation performance of ANN using RSM.

RSM					Confirmation runs
Fold	HN	TC	LR	MAE (validation)	Avg MA (validation)E	SD of MAE (validation)	95% PI half-width	PI coverage
1	4	802	0.005095	1.012	1.2255	0.0567	0.1422	No
2	8	194	0.005095	1.087	1.1364	0.0503	0.1262	Yes
3	4	951	0.005095	1.034	1.0781	0.0263	0.066	Yes
4	5	258	0.005095	1.095	1.1179	0.0518	0.1299	Yes
5	10	60	0.005095	1.035	1.1535	0.0611	0.1532	Yes
6	9	10	0.005095	0.98	1.0818	0.0533	0.1337	Yes
7	5	109	0.005095	1.169	1.2794	0.0675	0.1693	Yes
8	8	456	0.005095	1.062	1.198	0.0643	0.1613	Yes
9	7	208	0.005095	1.051	1.1091	0.0348	0.0873	Yes
10	4	109	0.005095	1.113	1.2119	0.0579	0.1452	Yes

Bold values indicate that this is the combination of hyperparameters that give the minimal average MAE (validation) among all folds.