Table 7. Results of the proposed methods compared to other methods from the literature on the USPS dataset.

Method		Reduced (ratio)	Accuracy	Training Speedup	Testing Speedup	Reduced SVs (ratio)
Whole dataset		1.00	0.922123	1.00	1.00	1.00
CBCH (k = 400) [15]		0.63182005	0.922073	1.621575	1.025486	0.961014
BPLSH (M = 30;L = 20) [16]		0.36004663	0.921923	3.254135	1.128958	0.857655
Shell Extraction (SE) [14]		0.4	0.918934	3.209440	1.118775	0.871665
Proposed Methods	DBI (r = 0.2)	0.2	0.911061	13.034276	1.785425	0.522033
	BRI (r = 0.1)	0.1	0.917389	27.714068	2.379853	0.387920
	BRIX (r = 0.04)	0.04	0.914898	62.172803	5.783704	0.138205
	SVO (k = 3)	0.15165272	0.913154	9.330182	1.209709	0.874003
	SVOX (k = 1)	0.07924839	0.916542	40.865746	4.393510	0.208468

The first row represents the metrics measured after training on the whole dataset. “Reduced (ratio)” is the ratio of the size of the resulting reduced dataset to the size of the whole dataset. “Reduced SVs (ratio)” is the ratio of the number of SVs of the reduced dataset to that of the whole dataset.