New Neighborhood Preserving Constraint Method for Support Vector Machines in SSS Classification
Support Vector Machines (SVMs) have solid theoretical foundations and excellent empirical success for supervised classification tasks. It tends to maximize the minimal between-class margin for good generalization ability. However, when introduced on small sample size training sets (e.g. SSS in medical diagnosis and image classification), where the number of samples is far less than the dimensionality of samples, the performance of SVMs classifier is not so desirable. In this paper, we propose a neighborhood preserving constraint based SVM method through balancing the margin separating distance metric and the local structure of the data manifold. This method enhances SVMs hi SSS case by explicitly preserving the relationship of with-class samples. It also can be deduced to a generalized framework of SVMs for efficient and uniform computation. The characteristics of the proposed method are further discovered in comparison to related methods. Experimental results on public data sets demonstrate that the proposed method has comparable or better generalization ability compared to some related SVMs algorithms.
Yanmin Niu Xuchu Wang
College of Computer and Information Science, Chongqing Normal University, Chongqing 400050, RR.China Key Laboratory of Optoelectronic Technology and Systems of Ministry of Education, Chongqing Universi
国际会议
重庆
英文
110-114
2011-01-21（万方平台首次上网日期，不代表论文的发表时间）