计算机工程与应用 ›› 2008, Vol. 44 ›› Issue (5): 177-179.

• 数据库与信息处理 • 上一篇    下一篇

快速的支持向量机多类分类研究

官 理1,祖 峰2,唐文胜1

  

  1. 1.湖南师范大学 计算机教学部,长沙 410081
    2.中国人民解放军 93320部分通信营,黑龙江 齐齐哈尔 161000
  • 收稿日期:2007-09-17 修回日期:2007-12-13 出版日期:2008-02-11 发布日期:2008-02-11
  • 通讯作者: 官 理

Research of fast multiclass SVM classification

GUAN Li1,ZU Feng2,TANG Wen-sheng1

  

  1. 1.Department of Computer Education,Hunan Normal University,Changsha 410081,China
    2.The Communication Batlation of 93320 Unit of the PLA,Qiqihar,Heilongjiang 161000,China
  • Received:2007-09-17 Revised:2007-12-13 Online:2008-02-11 Published:2008-02-11
  • Contact: GUAN Li

摘要: 研究了支持向量机多类算法DAGSVM(Direct Acyclic Graph SVM)的速度优势,提出了结合DAGSVM和简化支持向量技术的一种快速支持向量机多类分类方法。该方法一方面减少了一次分类所需的两类支持向量机的数量,另一方面减少了支持向量的数量。实验采用UCI和Statlog数据库的多类数据,并和四种多类方法进行比较,结果表明该方法能有效地加快分类速度。

关键词: 模式分类, 支持向量机, 多类分类, 计算复杂度

Abstract: The test speed advantage of the multiclass SVM method,DAGSVM(direct acyclic graph SVM),is analyzed.A fast multiclass SVM method,which combines the DAGSVM and the simplifying support vector solution,is proposed.The proposed method not only reduces the number of binary SVMs for one classification,but also reduces the number of support vectors.Experimented on UCI and Statlog multiclass data sets,compared with four other multiclass methods,results demonstrate that the proposed method can achieve a fast classification effectively.

Key words: pattern classification, Support Vector Machine, multiclass classification, computational complexity