一种基于混合二叉树结构的多类支持向量机分类算法

为提高多类支持向量机的分类效率,提出了一种基于混合二叉树结构的多类支持向量机分类算法。该混合二叉树中的每个内部结点对应一个分割超平面,该超平面通过计算两个距离最远的类的质心而获得,即该超平面为连接两质心线段的垂直平分线。每个终端结点(即决策结点)对应一个支持向量机,它的训练集不再是质心而是两类(组)样本集。该分类模型通常是超平面和支持向量机的混合结构,其中超平面实现训练早期的近似划分,以提升分类速度；而支持向量机完成最终的精确分类,以保证分类精度。实验结果表明,相比于经典的多类支持向量机方法,该算法在保证分类精度的前提下,能够有效缩短计算时间,提升分类效率。

Multi-class Classification Algorithm for SVM Based on Hybrid Binary Tree Structure

In order to improve the classification efficiency of mutli-class support vector mechine,a multi-class classification algorithm for support vector machine(SVM) based on hybrid binary tree structure was proposed.In the structure,each internal node corresponds to a partition hyperplane,which is obtained as perpendicular bisectors of linking two centroid segements of the two farthest classes from each other.Each terminal node(i.e.,decision node) is associated with a SVM,whose training set is two sets of samples instead of two centroids.In general,the resulting classification model represents a hybrid form,consisting of hyperplanes and SVMs.The approximate hyperplanes by centroids can provide fast partition in the early stages of the training phase,whereas the SVMs will perform the final precise decision.Experimental results show that compared with the classical multi-class SVM,the proposed algorithm can reduce the computational time and improve the classification efficiency with similar classification accuracy.