Classification method based on global and local support vector machine

Abstract

In order to solve the excessive consumption in space and time of standard support vector machine (SVM) and local SVM, a novel classification model called global and local SVM (GLSVM) is proposed. This new model obtains the global SVM by training non-boundary samples set. It obtains local SVM by training k-nearest neighbors in boundary samples set for each testing sample. In testing stage, the class of some testing samples is determined directly through global decision boundary. And the class of the others is determined with local SVM. Experiments prove that our proposed classification model has perfect performance in accuracy and efficiency.