SGB-ELM: An Advanced Stochastic Gradient Boosting-Based Ensemble Scheme for Extreme Learning Machine

Table 2

The comparison results between SGB-ELM and other representative algorithms on 4 regression datasets.

Dataset	Algorithm	Training time	Training RMSE (Dev)	Testing RMSE (Dev)	Hidden nodes	Iterations
Laser; ()	ELM	0.0097	12.3791 0.6067	12.7783 1.5863	80	N/A
	Simple ensemble	0.1547	12.1216 0.4585	12.8794 1.5415	80	10
	Bagging	0.7991	12.1707 0.5094	13.4085 1.5998	80	50
	Adaboost	0.1591	11.4460 0.4875	12.0666 1.2080	50	Max = 50
	SGB-ELM	3.4853	7.6354 0.3919	8.4170 1.0788	50 ()	50
Friedman; ()	ELM	0.0141	1.4220 0.0532	1.5124 0.0906	100	N/A
	Simple ensemble	0.2081	1.4005 0.0240	1.4791 0.0747	100	10
	Bagging	1.0144	1.4111 0.0197	1.4906 0.0701	100	50
	Adaboost	0.5219	1.2551 0.0304	1.3342 0.0587	60	Max = 50
	SGB-ELM	4.8853	1.0627 0.0136	1.1581 0.0346	60 ()	50
Mortgage; ()	ELM	0.0200	0.0855 0.0022	0.0961 0.0078	150	N/A
	Simple ensemble	0.3044	0.0843 0.0021	0.0947 0.0085	150	10
	Bagging	1.4544	0.0834 0.0018	0.0937 0.0077	150	50
	Adaboost	0.4778	0.0785 0.0020	0.0885 0.0058	80	Max = 50
	SGB-ELM	6.2434	0.0607 0.0015	0.0759 0.0056	80 ()	50
Wizmir; ()	ELM	0.0128	1.0906 0.0277	1.1263 0.0667	100	N/A
	Simple ensemble	0.2066	1.0869 0.0269	1.1203 0.0629	100	10
	Bagging	1.0366	1.0859 0.0259	1.1165 0.0620	100	50
	Adaboost	0.4331	1.0622 0.0519	1.1091 0.0857	60	Max = 50
	SGB-ELM	5.6525	1.0148 0.0258	1.1032 0.0615	60 ()	50