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

Table 4

Best hyperparameter setting and the test performance of DBN.

GS								Confirmation runs
Fold	HN	ni	n	LRRBM	LR	Nb	MAE (validation)	Avg MAE (validation)	SD of MAE (validation)	95% PI half-width	PI coverage
1	10	1000	505	0.1	0.05005	1025	0.987	1.0369	0.0318	0.0832	Yes
2	10	1000	10	0.05005	0.1	1025	1.0906	1.119	0.0383	0.1002	Yes
3	10	1000	1000	0.05005	0.05005	350	0.9718	1.01	0.0219	0.0573	Yes
4	10	505	505	0.1	0.05005	1025	1.0092	1.0647	0.0352	0.0921	Yes
5	10	505	1000	0.1	0.05005	1700	0.9978	1.0462	0.0456	0.1193	Yes
6	10	1000	10	0.1	0.1	1025	0.9313	1.0063	0.0246	0.0644	No
7	10	505	1000	0.1	0.1	1025	1.112	1.1737	0.038	0.0994	Yes
8	10	505	505	0.1	0.05005	1025	1.0361	1.0736	0.0333	0.0871	Yes
9	10	505	1000	0.1	0.1	1025	0.9893	1.0675	0.0277	0.0725	No
10	5.5	1000	505	0.05005	0.05005	350	1.0446	1.1091	0.0251	0.0657	Yes