An Optimized Neural Network Classification Method Based on Kernel Holistic Learning and Division

Table 2

Performance comparison on benchmark small sample datasets.

Datasets	Methods	Training time (s)	Testing		No. of kernels
			Blood	SVM		5.74	77.27	75.53	265
KHLD + SVM	13.85	77.92		75.36	113,265^[a]
ELM	0.01	76.48		75.32	80
KHLD + ELM	2.69	76.83		75.51	113,80^[a]
HSARBF-ELM	4.21	79.12		77.69	113,60^[a]
KHLD + HSARBF-ELM	7.39	79.85		78.14
CO-BP	3.85	72.13		71.28	8
KHLD + CO-BP	6.83	73.89		73.34	113,8^[a]
FOGDM-RBF	5.69	77.19		75.37	30
KHLD + FOGDM-RBF	7.87	78.35		76.92	113,30^[a]
Climate	SVM	1.86	92.32	92.64	49
	KHLD + SVM	3.74	92.86	92.82	13,49
	ELM	0.02	91.85	91.53	50
	KHLD + ELM	0.86	92.13	91.79	13,50
	HSARBF-ELM	2.78	93.47	92.41	13,50
	KHLD + HSARBF-ELM	3.82	94.21	93.13
	CO-BP	1.73	92.80	92.38	8
	KHLD + CO-BP	2.53	93.68	92.76	13,8
	FOGDM-RBF	2.11	92.26	92.07	8
	KHLD + FOGDM-RBF	3.20	93.39	92.82	13,8
HD	SVM	0.15	81.70	85.85	42
	KHLD + SVM	0.28	82.58	85.76	12,42
	ELM	0	79.56	78.62	30
	KHLD + ELM	0.05	80.41	79.83	12,30
	HSARBF-ELM	0.35	83.13	83.64	12,20
	KHLD + HSARBF-ELM	0.67	84.32	84.39
	CO-BP	0.08	80.58	80.37	5
	KHLD + CO-BP	0.23	81.52	81.26	12,5
	FOGDM-RBF	0.13	82.74	82.18	10
	KHLD + FOGDM-RBF	0.27	83.56	82.90	12,10
Sonar	SVM	0.12	80.85	84.97	46
	KHLD + SVM	0.26	82.56	85.34	49,46
	ELM	0	70.37	70.06	50
	KHLD + ELM	0.13	72.92	72.73	49,50
	HSARBF-ELM	0.64	76.74	76.09	49,40
	KHLD + HSARBF-ELM	0.95	79.59	80.31
	CO-BP	0.14	68.32	66.86	5
	KHLD + CO-BP	0.36	70.31	69.25	49,5
	FOGDM-RBF	0.27	73.63	72.58	16
	KHLD + FOGDM-RBF	0.62	74.82	74.29	49,16
SH	SVM	0.18	70.05	72.41	78
	KHLD + SVM	0.76	72.59	73.61	14,78
	ELM	0	66.24	65.56	50
	KHLD + ELM	0.23	67.92	66.72	14,50
	HSARBF-ELM	0.41	67.58	65.84	14,40
	KHLD + HSARBF-ELM	0.71	71.24	70.38
	CO-BP	0.12	71.26	70.68	7
	KHLD + CO-BP	0.38	73.18	72.80	14,7
	FOGDM-RBF	0.45	68.61	67.37	12
	KHLD + FOGDM-RBF	0.82	70.65	70.13	14,12
IS	SVM	7.46	90.56	—	96
	KHLD + SVM	20.53	91.84	—	32,96
	ELM	0.03	90.31	—	100
	KHLD + ELM	1.68	90.82	—	32,100
	HSARBF-ELM	5.17	92.23	—	32,80
	KHLD + HSARBF-ELM	7.23	93.17	—
	CO-BP	1.52	90.74	—	8
	KHLD + CO-BP	2.67	91.85	—	32,8
	FOGDM-RBF	4.63	89.58	—	36
	KHLD + FOGDM-RBF	6.39	90.54	—	32,36
Forests	SVM	1.86	72.19	74.51	125
	KHLD + SVM	2.94	73.60	74.78	51,125
	ELM	0.01	68.56	67.63	60
	KHLD + ELM	1.49	69.32	68.52	51,60
	HSARBF-ELM	1.83	69.93	69.37	51,50
	KHLD + HSARBF-ELM	3.35	71.65	71.32
	CO-BP	1.27	63.79	62.23	9
	KHLD + CO-BP	1.85	64.16	62.58	51,9
	FOGDM-RBF	2.32	69.24	68.87	30
	KHLD + FOGDM-RBF	3.94	70.62	70.42	51,30

^[a]The number of generated RBF kernels and the number of kernels/support vectors in each classifier.