中文摘要：

支持向量机决策树的精度和速度取决于树结构。为了获得好的泛化性能,应由可分性强的类为树的上层结点定义分类子任务。提出了一种新的支持向量机决策树设计算法。决策树中每个结点的分类子任务定义规则如下：采用模糊核C-均值将当前训练集粗分为两个子集,然后基于隶属度从各个子集中选择可分性强的子类定义当前结点的分类子任务,并将可分性弱的子类移至下层结点。实验结果表明,该方法的精度和速度都优于其他传统的多类分类方法。

英文摘要：

The accuracy and speed of decision-tree-based Support Vector Machine depend on the tree structure.To achieve high performance,classes with strong separability should be utilized to define classification task for the upper nodes of a decision tree.In this paper,a new algorithm for designing SVM with decision tree architecture was proposed.The classification task of each node was defined as follows： First,a coarse partition was applied to the current training set to generate two subsets by the algorithm of fuzzy kernel C-means.Then,according to membership degree,the sub-classes with great separability were selected from the two subsets to define classification task for the current node,while the sub-classes with weak separability were shifted to the lower nodes.Experimental results show that the proposed method is superior to other traditional multi-class classification methods in terms of both accuracy and speed.