基于非线性数据变换的离群点检测算法  

Outliers detection algorithm based on nonlinear data transformation

作　　者：徐雪松[1] 张谞[1] 宋东明[1] 张宏[1] 刘凤玉[1] 

机构地区：[1]南京理工大学计算机科学与技术学院,南京210094

出　　处：《中国工程科学》2008年第9期74-78,共5页Strategic Study of CAE

基　　金：国家自然科学基金资助项目(60273035)

摘　　要：为了提高高维数据集合离群数据挖掘效率,在分析了传统的离群数据挖掘算法优点和缺点的基础上,提出了一种离群点检测算法,首先将非线性问题转化为高维特征空间中的线性问题,然后利用非线性数据变换进行维数约减,对所得数据对象每个投影分量逐个判断数据点是否是离群点,通过实验证明该算法不仅可用于线性可分数据集的离群点检测,而且可用于线性不可分数据集的离群点检测,表明了算法的优越性。The data dimension reduction is the main method that can enhance the oufliers mining efficiency based on higher-dimension data set. A novel outlier detection algorithm is proposed after analyzing the advantages and disadvantages of the classical outlier mining algorithm in the paper. In this paper, we can transform nonlinear large-scale data into linear data in the feature space, and introduce a nonlinear data transformation to reduce data dimension. On the basis of each resulting vector, it determins whether the data is outlier data or not one by one. This paper shows that the algorithm is not only used to detect linear separable outlier data, but also used to detect nonlinear inseparable outlier data. This indicate that the algorithm has its obvious superiority.

关 键 词：维数消减 核函数 主成分 离群数据 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]