基于ε-Pinball损失函数的拉普拉斯双支持向量机  

Laplacian Twin Support Vector Machine Based on ε-Pinball Loss Function

作　　者：赵瑞卿 张晓丹[1] 赵伟峰 ZHAO Ruiqing;ZHANG Xiaodan;ZHAO Weifeng(School of Mathematics and Physics,University of Science and Technology Beijing,Beijing 100083,China)

机构地区：[1]北京科技大学数理学院

出　　处：《济南大学学报（自然科学版）》2019年第5期417-424,共8页Journal of University of Jinan(Science and Technology)

基　　金：国家自然科学基金青年科学基金项目(11801030)

摘　　要：考虑到拉普拉斯双支持向量机中的平方损失函数对分类超平面两侧的同类样本点给予了相同重视,当出现噪声或离群点时,所得分类超平面可能会出现偏离的现象,为了减小噪声或离群点的影响,提出基于ε-Pinball损失函数的拉普拉斯双支持向量机;给出正、负损失的概念,探讨参数τ对分类超平面的影响,分析参数ν的意义,并进行数值实验。结果表明,通过调节参数τ,可增强模型的灵活性,使得模型具有较好的分类能力及抗噪性。Given that the square loss function in Laplacian twin support vector machine gave the equivalent attention to sample points of the same kind,when there were noises or outliers,the classification hyperplane obtained by using this algorithm might be biased,and a new Laplacian twin support vector machine was proposed based on ε-Pinball loss function to effectively handle this case.The concepts of positive loss and negative loss were given,the effect of parameterτon classification hyperplane was discussed,and the property of parameterνwas analyzed.Numerical experiments on eight benchmark datasets were performed.The results indicate that the flexibility of the model can be enhanced by adjusting the parameterτ,and the model has better classification ability and noise resistance.

关 键 词：拉普拉斯双支持向量机 平方损失函数 ε-Pinball损失函数 正损失 负损失 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]