计算机工程与应用 ›› 2014, Vol. 50 ›› Issue (14): 127-130.

• 数据库、数据挖掘、机器学习 • 上一篇    下一篇

鲁棒双子支持向量回归

孙少超,应忠于,李伟春,胡云琴,朱  麟   

  1. 公安海警学院,浙江 宁波 315801
  • 出版日期:2014-07-15 发布日期:2014-08-04

Robust twins support vector machine for regression

SUN Shaochao, YING Zhongyu, LI Weichun, HU Yunqin, ZHU Lin   

  1. China Maritime Police Academy, Ningbo, Zhejiang 315801, China
  • Online:2014-07-15 Published:2014-08-04

摘要: 针对传统支持向量回归机缺乏鲁棒性而鲁棒支持向量回归机稀疏性不理想,提出了新的支持向量回归方法(鲁棒双子支持向量回归)。为了求解的方便,该方法的损失函数由两个可微的凸函数构成,并且采用CCCP技术对其进行求解。该方法在获得良好稀疏性的同时有效地抑制了过失误差的影响。通过人工数据和现实真实数据对该方法的测试,验证了新方法的有效性。

关键词: 支持向量回归, 稀疏性, 鲁棒, 损失函数

Abstract: A Robust Twins Support Vector Regression(RTSVR) is proposed to overcome the weakness of the standard support vector regression and robust support vector regression. Its novel robust loss function which owns advantage in robustness and sparseness property for support vector regression is proposed. To improve the computational efficiency, it is constructed by combining two differentiable convex functions. Then the concave-convex procedure is used to solve the RTSVR by transforming the non-convex problem into a sequence of convex ones. The RTSVR can not only obtain better sparseness property but also restrain the outliers of training samples. Experiments have been done on artificial and benchmark datasets and the results show the effectiveness of the proposed RTSVR.

Key words: support vector regression, sparseness, robustness, loss function