基于SBA和KNN的多类分类算法  

Multi-class classification algorithm based on SBA and KNN

作　　者：陈松贵[1] 李峰[1] 王飞燕[1] 

机构地区：[1]长沙理工大学计算机与通信工程学院,湖南长沙410076

出　　处：《计算机工程与设计》2008年第22期5786-5788,5791,共4页Computer Engineering and Design

基　　金：湖南省自然科学基金项目(05JJ30123);湖南省教育厅科学研究基金项目(05C246)

摘　　要：采用二叉树结构对多个二值支持向量机子分类器组合,可实现多类问题的分类,并且还可克服传统(SVM)多类算法存在的不可分区域的情况。为了加快大规模数据样本分类时的训练速度以及减少分类的时间复杂度,将一种新的二分类算法——SBA应用于改进的BTS-KNN算法中,可以构造出更加平衡的分类二叉树以减少分类次数。实验结果表明,在进行大规模样本数据分类时,该算法在保证分类精度的情况下分类速度有明显提高。The method constructing and combining several binary SVMs （support vector machines） with a binary tree can solve multiclass problems, and resolve the unclassifiable regions that exist in the conventional muliclass SVMs. In order to improve the efficiency of classification with large numbers, a new binary classification algorithm SBA is introduced at first, then it is applied to improved BTSKNN algorithm, it will make the binary SVM tree be balance to reduce the degree of classification. The experiments show that maintaining comparable accuracy, the algorithm is much faster to be trained than other methods when in classification which has large numbers.

关 键 词：支持向量机 序贯最小化算法 二叉树支持向量机 K-近邻 SBA 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]