An Improved Binary Tree SVM Classification Algorithm Based on Bayesian

Due to the great generalization and the support of statistics, support vector machines (SVMs) has been widely applied to resolving multi-class classification problem. Numbers of multi-class SVM have been proposed. Compared with other multi-class SVM, binary tree of SVM (BTS) takes a good advantage of lower time consuming. However, there is some unnecessary data reassignment during constructing a binary tree, which makes BTS can´t resolve the high-dimensional multi-class classification problem accurately. In this paper, a bayesian-based BTS classification algorithm (b-BTS) has been proposed. Experiments demonstrated that b-BTS is superior to BTS in resolving classification problem, such as image classification problem.