电子学报 ›› 2017, Vol. 45 ›› Issue (2): 408-416.DOI: 10.3969/j.issn.0372-2112.2017.02.020

• 学术论文 • 上一篇    下一篇

改进的投影孪生支持向量机

陈素根1,2, 吴小俊1   

  1. 1. 江南大学物联网工程学院, 江苏无锡 214122;
    2. 安庆师范大学数学与计算科学学院, 安徽安庆 246133
  • 收稿日期:2015-09-22 修回日期:2015-11-17 出版日期:2017-02-25 发布日期:2017-02-25
  • 通讯作者: 吴小俊
  • 作者简介:陈素根,男,1982生,江南大学物联网工程学院博士研究生,安庆师范大学副教授.研究方向为模式识别与智能系统、机器学习等.E-mail:chensugen@126.com
  • 基金资助:

    国家自然科学基金(No.61373055);安徽省高等学校自然科学研究重点项目(No.KJ2015A266,No.KJ2016A431)

Improved Projection Twin Support Vector Machine

CHEN Su-gen1,2, WU Xiao-jun1   

  1. 1. School of Iot Engineering, Jiangnan University, Wuxi, Jiangsu 214122, China;
    2. School of Mathematics and Computational Science, Anqing Normal University, Anqing, Anhui 246133, China
  • Received:2015-09-22 Revised:2015-11-17 Online:2017-02-25 Published:2017-02-25

摘要:

针对投影孪生支持向量机(Projection Twin Support Vector Machine,PTSVM)在训练和求解过程中存在的问题,提出了一类改进的投影孪生支持向量机(Improved PTSVM),简称为IPTSVM.该文首先构造了改进的线性投影孪生支持向量机,然后利用核技巧轻松将其推广到了非线性形式.本文的主要贡献有:(1)提出了投影孪生支持向量机的新模型,克服了原始PTSVM在训练之前需要求解两个逆矩阵的问题;(2)继承了传统SVM(Support Vector Machine)的精髓,利用核技巧直接将线性IPTSVM推广到非线性形式;(3)引入了一个新的参数,可以调节模型的性能,提高了IPTSVM的分类精度.实验结果表明,与PTSVM算法相比较,IPTSVM不仅提高了分类精度,而且克服了PTSVM的一些不足.

关键词: 支持向量机, 非平行平面支持向量机, 投影孪生支持向量机, 模式分类

Abstract:

An Improved Projection Twin Support Vector Machine (IPTSVM) is presented.The target of the proposed IPTSVM is to deal with a set of problems in the training and solving steps of PTSVM.We first propose a linear IPTSVM for binary classification.Then we extend it to the corresponding nonlinear version using kernel tricks.The paper has three main contributions to the community:(1) A new PTSVM-based method is proposed,in which we do not have to compute the inverse of a large matrix before the training step.(2) We design the nonlinear IPSVM that is obtained by using kernel tricks.(3) A new parameter is introduced,which can adjust the performance of the model and improve the classification accuracy of IPTSVM.Experimental results obtained from several datasets demonstrate that,compared with PTSVM,IPTSVM not only improves the classification accuracy but also overcomes some deficiencies to a certain extent.

Key words: support vector machine, nonparallel hyperplane support vector machine, projection twin support vector machine, pattern classification

中图分类号: