摘要
针对支持向量机理论中现存的问题:多类分类问题和对于噪音数据的敏感性,提出了一种模糊多类支持向量机算法.该算法是在Weston等人提出的多类SVM分类器的直接构造方法中引入模糊成员函数,针对每个输入数据对分类结果的不同影响,该模糊成员函数得到相应的值,由此可以得到不同的惩罚值,并且在构造分类超平面时,可以忽略那些对分类结果影响很小的数据.在充分的数值实验基础上,将文中提出的方法应用于当前一个重要的应用领域———计算机网络入侵检测问题,并得到了较好的实验结果.理论分析与数值实验都表明,该算法是切实可行的,并具有良好的鲁棒性.
SVMs (support vector machines) are very powerful in classification and regression, and have been applied to many application fields successfully, such as pattern recognition and data mining. But there are two kinds of problems to be solved in such filed, one is the multi-class classification problem, and the other is the sensitivity to the noisy data. In order to overcome thesis difficulties, an approach of fuzzy multi-class Support Vector Machine is proposed in this paper, which is refered as FMSVM in the present paper. In many cases, each input point may not be fully assigned to one of the classes. This paper introduces a fuzzy membership function to the penalty in the quadratic problem of proposed by Weston and Watkins, the membership function acquire different values for each input data according to their different affects on the classification result. Hence, the data which affect the classification result a little is ignored. Therefore different input points can make different contributions to the learning of the decision surface-the optimal separating hyper-plane. The experiments contain two parts, i.e., the numerical experiments and the experiment on intrusion detection.
出处
《计算机学报》
EI
CSCD
北大核心
2005年第2期274-280,共7页
Chinese Journal of Computers
基金
国家"十五"重点科技攻关项目基金(2002BA407B)资助