The multistage support vector machine

Abstract

The support vector machine (SVM) was originally designed for binary classification. How to effectively extend it for multiclass classification is still an ongoing research issue. There exist several methods to construct a multiclass classifier by combing several binary classifiers, such as "one-against-one", "one-against-all" and directed acyclic graph SVM. In the paper we give a new method to combine several binary classifiers other than the above three methods. Our idea is to cluster the samples into two classes, and then classify them. Based on this idea, we present the concept of multistage support vector machine. A comparison between our proposed method with the other three methods is conducted on the Iris database. Comparative results show that our proposed method has a better performance than the other three methods.