一种基于几何分布的新支持向量机多分类方法

二叉树支持向量机是多分类问题的一种有效方法，然而分类的效果与二叉树的结构密切相关。获得更好的分类效果和更高的效率，要使得二叉树高度尽量小而两个子类尽量易分。距离通常用来衡量两个类的分离程度，但不能反映类的分布情况。考虑到多分类中类的分布，文中定义新的分离度和相似度来衡量两个类的分离度，并且提出了一中新的基于几何分布二叉树支持向量机多分类算法，该方法使得二叉树高度尽量小而两个子类尽量易分。实验表明该方法具有较高的分类准确率和效率。

A New SVM Multi-classification Method Based on Geometric Distribution

BTSVM is an effective approach for solving multi-class problems. However, the classification performance of the classifier is closely related to the tree structure. In order to get better performance and higher efficiency,it is necessary to make the two sub-classes more separable and make binary tree less hierarehicals. Distance measure is common used as a separability measure between classes,but it does not reflect the distribution of the classes. In consideration of distribution, a new separability measure and a new similarity measure are defined to measure the separability and similarity between classes. Moreover,a novel geometric-distribution-based BTSVM is proposed to make the two sub-classes more separable and binary tree less hierarchicals. Experiments made on three data sets prove the high effi- ciency and good performance of the proposed algorithm.