摘要
针对传统的支持向量机(SVM)对训练样本中的噪声和野值特别敏感而导致的过学习问题,文中提出了一种新的基于动态核函数的模糊支持向量机(FSVM)。该方法不仅考虑了样本点到类中心的距离,而且还考虑了样本间的密切度,结合这两种思想在特征空间中构造了一种新的基于动态核函数的模糊隶属度。仿真实验表明,该方法有较好的分类精度和推广能力并且在理论上具有一般性和能够有效地减弱野值的影响。
Since traditional support vector machine is very sensitive to the noises and outliers in the training samples,so it is easy to produce the over-fitting problem.To solve the problem,fuzzy support vector machine is introduced based on the dynamic kernel method in this paper.The fuzzy membership is defined not merely by the distance between a point and its class center,but also by two different points of the sample,which is depicted as the affinity between them.The experimental simulations show that FSVM with the new membership function not only has better classification accuracy and generalization,but also has universality in theory and gives good performance on reducing the effects of outliers.
出处
《南京邮电大学学报（自然科学版）》
2010年第6期43-47,共5页
Journal of Nanjing University of Posts and Telecommunications：Natural Science Edition
基金
国家自然科学基金(10371106
10471114
61070234
61071167)
江苏省高校自然科学基金(04KJB110097
08KJB520003)
南京邮电大学攀登计划(NY207064)资助项目
关键词
模糊支持向量机
模糊隶属度函数
动态核函数
分类
信息几何
fuzzy support vector machine
fuzzy membership function
dynamic kernel function
classification
information geometry
