改进的二叉树支持向量机在多分类中的应用  

Application of Improved Binary Tree Support Vector Machine in Multi-classification

作　　者：李燕玲[1] 苏一丹[1] 

机构地区：[1]广西大学计算机与电子信息学院,广西南宁530004

出　　处：《计算机技术与发展》2014年第7期181-184,共4页Computer Technology and Development

基　　金：教育部人文社会科学研究项目(11YJAZH080)

摘　　要：在多分类问题中,分类算法的优劣直接影响到最终分类结果的好坏。现有的多分类算法中,基于支持向量机的多分类算法在综合性能方面要优于其他算法,但是,这些较优算法同样面临一些多分类中常见的问题,如不可分问题和效率低问题。针对这些问题,文中提出了一种改进的二叉树支持向量机多分类算法,该算法综合考虑了两个类之间的距离和分布情况对可分离性的影响,并采用最容易分离的类最先分割出来的策略来建立树的结构。通过在不同的数据集上进行测试,表明该方法不仅解决了多分类的不可分问题,还能提高分类的效率和准确度,可更好地解决现实中的多分类问题。The quality of algorithms has directly impacted on the final classification results in multi-class classification. In current algo-rithms for multi-classification,those which are based on Support Vector Machine （ SVM） have better comprehensive performance than others. But they also face some common problems,such as unclassifiable regions and low efficiency. For these problems,a modified bina-ry tree SVM multi-classification algorithm which is based on the effect of distance and distribution of classes to inter-class separability is proposed,using the strategy of the easiest class to separate for the first partition to establish the structure of the tree. Tests in different data sets show that this method can not only solve the unclassifiable regions,but also can improve the efficiency and accuracy of classification.

关 键 词：支持向量机 多分类 二叉树 超球体 

分 类 号：TP39[自动化与计算机技术—计算机应用技术]