基于隶属度分离测度SVM决策树层次结构设计方法

Method of designing hierarchical structure of SVM decision tree based on separating measure with degree of membership

从样本的类空间分布和随机测试样本对每个类别的隶属度两方面考虑,对现有的分离测度进行了改进,并给出了一种基于隶属度分离测度的SVM决策树多类分类算法。实验表明,对于随机测试样本属于每个类别的概率均不相同的多类分类问题,基于隶属度分离测度的SVM决策树在与传统的SVM决策树有着基本相同的分类精度情况下,具有更快的分类速度。

This paper proposed a new separating measure, and designed the hierarchical structure of SVM decision tree based on this new separating measure. The experimental results show that when the random testing sample has different degree of membership to different classes, this SVM decision tree algorithm with the new separating measure speeds up the separation under the same separation ability.