作者投稿和查稿 主编审稿 专家审稿 编委审稿 远程编辑

计算机工程 ›› 2012, Vol. 38 ›› Issue (24): 179-181. doi: 10.3969/j.issn.1000-3428.2012.24.042

• 人工智能及识别技术 • 上一篇    下一篇

一种双重正则化支持向量机的改进算法

秦传东 1,2,刘三阳 2   

  1. (1. 北方民族大学信息与计算科学学院,银川 750021;2. 西安电子科技大学理学院,西安 710071)
  • 收稿日期:2011-09-05 修回日期:2011-11-28 出版日期:2012-12-20 发布日期:2012-12-18
  • 作者简介:秦传东(1976-),男,博士,主研方向:数据挖掘,最优化理论与应用;刘三阳,教授、博士、博士生导师
  • 基金资助:

    国家自然科学基金资助项目(60974082);国家自然科学基金青年基金资助项目(10901004)

An Improvement Algorithm of Doubly Regularized Support Vector Machine

QIN Chuan-dong 1,2, LIU San-yang 2   

  1. (1. School of Information and Computation Science, North University for Ethnics, Yinchuan 750021, China; 2. School of Science, Xidian University, Xi’an 710071, China)
  • Received:2011-09-05 Revised:2011-11-28 Online:2012-12-20 Published:2012-12-18

摘要: 针对L1范数支持向量机和L2范数支持向量机在分析部分小样本、高维数、变量高相关的数据时效果不理想的问题,在综合利用这2种支持向量机优点的基础上,提出一种双重正则化支持向量机的改进算法。通过正号函数和二次多项式损失函数将问题转化为可微的无条件约束优化问题,便于采用多种优化算法进行运算。实验结果证明,该改进算法可取得较好的分类准确率。

关键词: L1范数支持向量机, L2范数支持向量机, 正号函数, 二次多项式函数, BFGS算法, 双重正则化

Abstract: When L1-norm support vector machine and L2-norm support vector machine are used to analyse the datasets with small sample, high dimension and high correlation in parts of the variables, the effects of them are not satisfactory. Taking the good advantages of the two methods, an improvement algorithm of doubly regularized support vector machine is proposed. But the inequality constraints and the non-differentiable norm bring many troubles. A positive function and a quadratic polynomial loss function are introduced to change the optimization problem into a differentiable and unconditional constraints one which is easy to compute using many optimization algorithms. Experimental results show the improvement gains better effects.

Key words: L1-norm support vector machine, L2-norm support vector machine, positive function, quadratic polynomial function, BFGS algorithm, doubly regularization

中图分类号: