Knowledge-based Support Vector Classification Based on C-SVC
Chunhua Zhang, Xiaojian Shao, Dewei Li · Procedia Computer Science · 2013
In many real-world classification problems, we are not only given the traditional training sets, but also provided by some prior information. This paper focuses on the classification problems in such scenario where prior knowledge is available in the form of multiple polyhedral sets. In the traditional knowledge-based support vector machines (KBSVMs) introduced by Mangasarian and his co-workers, prior knowledge was incorporated into 1-norm SVC. However, these methods lost the superiority of the standard SVC (C-SVC) because they are based on 1-norm SVC. In this paper, the new KBSVC based on C-SVC are proposed in an intuitive way. After transforming the prior knowledge into linear constraints of the quadratical programming of C-SVC, we derive two models: the linear and nonlinear KBSVC, which corresponds to the linear and the nonlinear C-SVC respectively. We believe that our new models can preserve the performance of C-SVC.