基于Pinball损失函数的半监督支持向量机  

Semi-supervised Support Vector Machines with Pinball Loss Function for Classification

作　　者：刘旭东[1] 张晓丹[1] 

机构地区：[1]北京科技大学数理学院,北京100083

出　　处：《济南大学学报（自然科学版）》2017年第1期42-47,54,共7页Journal of University of Jinan(Science and Technology)

基　　金：中央高校基本科研业务费资助项目(FRF-BR-12-021)

摘　　要：基于Pinball损失函数的流形正则化半监督二分类支持向量机,建立一个新的半监督二分类支持向量机模型,采用对偶方法导出新模型的对偶问题,给出求解模型的算法步骤,并对模型进行了参数分析;针对新模型的分类准确性和抗噪声性能进行数值实验。结果表明,合适的损失函数参数有助于提升模型的分类效果,并且新模型具备一定的抗噪声能力和对再采样的稳定性。Based on the Pinball loss function, a binary classification manifold regularization semi-supervised support vector machine was studied and a new binary classification semi-supervised support vector machine was constructed. Using the dual method, the dual problem of the original new model was developed and the parameter of the model was analyzed. Numerical experiments related with classification accuracy and anti-noise were performed. The results show that suitable parameters of Pinball loss function help to improve the classification performance and that the model is noise insensitive and re-sample stable.

关 键 词：半监督 损失函数 流形正则化 支持向量机 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]