基于迭代方法的模糊孪生支持向量机  

Fuzzy twin support vector machine based on iterative method

作　　者：李凯[1] 胡少方 王亮[1] 董华[1] 许建忠[2] 

机构地区：[1]河北大学计算机科学与技术学院,河北保定071002 [2]河北大学物理科学与技术学院,河北保定071002

出　　处：《河北大学学报（自然科学版）》2015年第4期399-404,共6页Journal of Hebei University(Natural Science Edition)

基　　金：国家自然科学基金资助项目(61375075)

摘　　要：模糊孪生支持向量机通过为每个训练样本赋予不同的模糊隶属度来构建2个最优非平行分类面,以便减少噪声或孤立点对非平行分类面的影响,进一步提高了支持向量机的性能.本文结合超松弛迭代法对模糊孪生支持向量机进行了研究,通过迭代技术求解凸二次规划问题中的拉格朗日乘子,减少了支持向量机的训练时间,在UCI标准数据集上分别对C-FTSVM和v-FTSVM进行了实验研究,并对使用传统求拉格朗日乘子的方法与超松弛迭代(SOR)的方法进行了对比,表明了使用超松弛迭代法不仅在时间性能上得到了提高,而且其分类正确率也优于传统的方法.Fuzzy twin support vector machine takes the importance of the training samples on the learning of the decision hyperplane into account with respect to the classification task.Two optimal nonparallel hyperplanes were builded to achieve classification accuracy by applying a fuzzy membership to each training sample to reduce the effects of the noises and outliers on the hyperplane.In this paper,over-relaxation iterative method was used to solve the optimization problem of the fuzzy twin support vector machine in order to obtain lagrange multipliers.Experiments were conducted for both C-FTSVM and v-FTSVM algorithms on UCI benchmark datasets and compared with the traditional method of soving lagrange multipliers.The experimental results indicated that the speed and accuracy of classification of using successive overrelaxation method（SOR）for fuzzy twin support vector machine were superior to those of using the traditional optimalization method.

关 键 词：模糊孪生支持向量机 超松弛迭代 分类 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]