Online Sequential Projection Vector Machine with Adaptive Data Mean Update

Table 12

Training and testing accuracy of PVM and OSPVM with the same hidden nodes.

Dataset	Algorithms	#Hidden nodes	Training accuracy	Testing accuracy
Face	OSPVM (1-by-1 and 16-by-16)	65	99.81%	92.30%
	Batch-PVM	65	99.81%	92.30%
Secom	OSPVM (1-by-1 and 16-by-16)	60	93.37%	93.35%
	Batch-PVM	60	93.37%	93.35%
Arcene	OSPVM (1-by-1 and 16-by-16)	85	94.63%	90.80%
	Batch-PVM	85	94.63%	90.80%
Dexter	OSPVM (1-by-1 and 16-by-16)	160	98.38%	91.25%
	Batch-PVM	160	98.38%	91.25%
Multi.fea.	OSPVM (1-by-1 and 16-by-16)	160	99.98%	95.67%
	Batch-PVM	160	99.98%	95.67%
News20	OSPVM (1-by-1 and 16-by-16)	1000	84.89%	83.12%
	Batch-PVM	1000	84.89%	83.12%
Sector	OSPVM (1-by-1 and 16-by-16)	160	87.98%	79.01%
	Batch-PVM	160	87.98%	79.01%