基于PCA及属性距离和的孤立点检测算法  

Algorithm for outlier detection based on principal component analysis and sum of attributes distance

作　　者：张忠平[1] 宋少英[1] 宋晓辉[1] 

机构地区：[1]燕山大学信息科学与工程学院,河北秦皇岛066004

出　　处：《计算机工程与应用》2009年第17期139-141,243,共4页Computer Engineering and Applications

基　　金：国家自然科学基金(No.60773100);教育部科学技术研究重点项目(No.205014);河北省教育厅科研计划项目(No.2006143)~~

摘　　要：提出了一种基于主分量分析和属性距离和的孤立点检测算法。该方法首先通过主分量分析方法从众多属性中提取出满足累计贡献率的主分量,同时利用PCA变换矩阵把原始数据集转换到由主分量组成的新的特征空间上,之后对转换后的数据集用属性距离和的方法对孤立点进行检测。实验结果证明了基于主分量分析和属性距离和的孤立点检测算法的有效性。An outlier detection algorithm based on principal component analysis and the sum of attributes distance is proposed. The algorithm firstly extracts the principal components from many attributes satisfying accumulative contribution rate.Simultaneously,by the PCA matrix original dataset is transformed to a new feature space composed of principal component.Then outliers are detected using the approach of the sum of attributes distance in the transformed datasets.The results of the experiment show that the outlier detection algorithm based on principal component analysis and the sum of attributes distance is effective.

关 键 词：孤立点 主分量分析 累计贡献率 属性距离和 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]