`a`
Journal of Industrial and Management Optimization (JIMO)
 

On regularisation parameter transformation of support vector machines

Pages: 403 - 415, Volume 5, Issue 2, May 2009      doi:10.3934/jimo.2009.5.403

 
       Abstract        Full Text (191.0K)       Related Articles       

Hong-Gunn Chew - School of Electrical and Electronic Engineering, The University of Adelaide, SA 5005, Australia (email)
Cheng-Chew Lim - School of Electrical and Electronic Engineering, The University of Adelaide, SA 5005, Australia (email)

Abstract: The Dual-nu Support Vector Machine (SVM) is an effective method in pattern recognition and target detection. It improves on the Dual-C SVM, and offers competitive performance in detection and computation with traditional classifiers. We show that the regularisation parameters Dual-nu and Dual-C can be set such that the same SVM solution is obtained. We present the process of determining the related parameters of one form from the solution of a trained SVM of the other form, and test the relationship with a digit recognition problem. The link between the Dual-nu and Dual-C parameters allows users to use Dual-nu for ease of training, and to switch between the two forms readily.

Keywords:  Support Vector Machine, Pattern recognition, Quadratic optimisation.
Mathematics Subject Classification:  Primary: 68T10; Secondary: 90C20.

Received: March 2008;      Revised: September 2008;      Available Online: April 2009.