电子学报 ›› 2014, Vol. 42 ›› Issue (5): 998-1003.DOI: 10.3969/j.issn.0372-2112.2014.05.026

• 科研通信 • 上一篇    下一篇

一种改进的最小二乘孪生支持向量机分类算法

储茂祥1,2, 王安娜1, 巩荣芬1,2   

  1. 1. 东北大学信息科学与工程学院, 辽宁沈阳110819;
    2. 辽宁科技大学电子与信息工程学院, 辽宁鞍山114051
  • 收稿日期:2013-02-26 修回日期:2013-08-21 出版日期:2014-05-25 发布日期:2014-05-25
  • 作者简介:储茂祥 男,1978年生于安徽桐城.东北大学信息科学与工程学院博士研究生.辽宁科技大学电子与信息工程学院讲师.研究方向为模式识别、智能系统. E-mail:chu52-2004@163.com王安娜 女,1956年生于辽宁鞍山.东北大学信息科学与工程学院博士生导师、教授.研究方向为模式识别、电力电子系统、复杂工业过程故障诊断.
  • 基金资助:
    国家自然科学基金(No.61050006)

Improvement on Least Squares Twin Support Vector Machine for Pattern Classification

CHU Mao-xiang1,2, WANG An-na1, GONG Rong-fen1,2   

  1. 1. College of Information Science and Engineering, Northeastern University, Shenyang, Liaoning 110819, China;
    2. School of Electronic and Information Engineering, University of Science and Technology Liaoning, Anshan, Liaoning 114051, China
  • Received:2013-02-26 Revised:2013-08-21 Online:2014-05-25 Published:2014-05-25

摘要: 提出了一种新的模式分类器,即广泛权重的最小二乘孪生支持向量机.该支持向量机在正、负两类样本上广泛地增加权重,很好地抑制了交叉噪声样本对数据分类的影响.其次,根据间隔最大化原理,该支持向量机在目标函数上增加了一个正规化项,实现结构风险最小化和避免在求解该目标函数时可能对病态矩阵求逆的处理.同时,提出了利用一种指数函数计算训练样本的密度来获得样本权重值的算法.该算法能够有效缩减计算权重的时间,且具有较强的鲁棒性.实验证明本文提出的广泛权重的最小二乘孪生支持向量机能够实现高精度和高效率的分类效果,而且特别适合于含有交叉噪声样本的数据集分类.

关键词: 模式分类, 最小二乘, 孪生支持向量机, 权重, 指数函数

Abstract: Widely weighted least squares twin support vector machine (WWLSTSVM) is proposed for pattern classification.In WWLSTSVM,weights are widely added on error variables of data samples both in one class and the other.This widely weighted method is especially effective on eliminating the interference of intercrossing noise samples.Moreover,a regularization term is added with the theory of maximizing margin,in which the structural risk is minimized and the possible ill-conditioning is avoided for matrix inversion.Also,an effective weight algorithm with exponential function is proposed to reduce the time complexity of computing weight values and enhance its robustness for cross plane dataset.Comparative experiments show that WWLSTSVM obtains better results on eliminating the interference of noise samples and higher classification accuracy with less computing time in both linear and nonlinear cases compared with the other classifiers.

Key words: pattern classification, least squares, twin support vector machine, weight, exponential function

中图分类号: