摘要
基于核学习理论提出一种方差保持的异常检测分类器(CP-ND)。使用正常类方差使分类线与正常类空间分布保持一致,最大化分类线和异常点之间的间隔,通过二次规划求解对偶问题。训练参数v、v1和v2之间有简单约束关系,vv1和vv2分别指示正常类和异常类的误分率上界及支持向量率下界。医学诊断数据集的测试结果表明,CP-ND具有较高的分类精度。
This paper presents an anomalous detection classifier of variance preserving named CP-ND.The covariance of normal examples is applied to preserve the statistical distribution of normal data and the margin between the decision hyper plane and abnormal points are maximized.The dual problem of this model can be solved as a quadratic programming.There are some inequalities among the three parameters of v,v1 and v2 introduced by this classifier,for normal class,vv1 indicates the rate of training misclassification and the rate of support vectors and vv2 indicates corresponding rates of abnormal class,those inequalities can be used to tune the three parameters.CP-ND classifier is evaluated on real-world medical diagnosis data sets.
出处
《计算机工程》
CAS
CSCD
北大核心
2011年第23期24-26,共3页
Computer Engineering
基金
国家自然科学基金资助项目(90820002)
中央高校基本科研业务费专项基金资助项目(JUDCF09034)
关键词
分类器
异常检测
方差保持
支持向量机
协方差矩阵
classifier
anomalous detection
variance preserving
Support Vector Machine（SVM）
covariance matrix
