摘要
为提高支持向量域分类器(SVDC)的分类精度和鲁棒性,提出基于K近邻(KNN)和支持向量域描述(SVDD)的分类器KNN-SVDD(KSVDD)。该分类器对单类内部的样本采用SVDD的判别准则,对类交叉区域及描述边界外的样本采用KNN的判别准则。通过拒绝描述边界外的样本,KSVDD可应用于拒识判别。UCI数据集上的数值实验表明,KSVDD分类精度与支持向量机(SVM)相当且均比SVDC高,训练时间比SVM短,鲁棒性强,在拒识判别中有良好表现。
To improve the accuracy and robustness of Support Vector Domain Classifier（SVDC）,KSVDD is proposed based on K-Nearest Neighbor（KNN） and Support Vector Domain Description（SVDD）.The classifier takes SVDD determination for test samples inside single class,and adopts the KNN rule for test samples inside the overlapped regions or outside the description boundaries.By rejecting samples outside the description boundaries,the classifier can also be generalized to rejection determination.Numerical experiments on UCI data show that KSVDD has higher accuracy over SVDC,is comparable with SVM,has lower training time than SVM,is more robust and has good performances in rejection determination.
出处
《计算机工程》
CAS
CSCD
北大核心
2010年第19期195-197,共3页
Computer Engineering
基金
西安统计研究院基金资助重点项目(09JD07)
关键词
支持向量域分类器
K近邻
支持向量域描述
拒识判别
鲁棒性
Support Vector Domain Classifier（SVDC）
K-Nearest Neighbor（KNN）
Support Vector Domain Description（SVDD）
rejection determination
robustness
