	US 7,490,071 B2
	Support vector machines processing system
Boriana L. Milenova, Somerville, Mass. (US); Joseph S. Yarmus, Groton, Mass. (US); Marcos M. Campos, Cambridge, Mass. (US); and Mark A. McCracken, Belmont, Mass. (US)
Assigned to Oracle Corporation, Redwood Shores, Calif. (US)
Filed on Aug. 27, 2004, as Appl. No. 10/927,111.
Claims priority of provisional application 60/498699, filed on Aug. 29, 2003.
Prior Publication US 2005/0049990 A1, Mar. 03, 2005
Int. Cl. G06F 17/00 (2006.01); G06N 5/00 (2006.01)

U.S. Cl. 706—45 [706/46; 706/12]

18 Claims

1. A computerized method for support vector machine processing comprising:

storing data;

providing an interface to client software;

building a support vector machine model on at least a portion of the stored data, based on a plurality of model-building parameters;

estimating values for at least some of the model-building parameters; and applying the support vector machine model using the stored data to generate a data mining output;

wherein the parameter estimation is performed by selecting a complexity parameter C method for a linear kernel method by:

selecting a random sample comprising k pairs of training vectors of opposite class, one member of each pair being of positive class and one member of each pair being of negative class;

for each pair, assigning C to have a value such that a margin for the positive class member of the pair is 1 and a margin for the negative class member of the pair is -1;

ordering the assigned values of C; and

selecting as C a k-th highest assigned value of C.