作者投稿和查稿 主编审稿 专家审稿 编委审稿 远程编辑

计算机工程 ›› 2013, Vol. 39 ›› Issue (2): 178-181. doi: 10.3969/j.issn.1000-3428.2013.02.036

• 人工智能及识别技术 • 上一篇    下一篇

基于小波密度估计的数据流离群点检测

刘耀宗 1,2,张 宏 1,孟 锦 1,2,韩法旺 2   

  1. (1. 南京理工大学计算机学院,南京 210094;2. 南京森林警察学院信息系,南京 210046)
  • 收稿日期:2012-03-16 修回日期:2012-05-31 出版日期:2013-02-15 发布日期:2013-02-13
  • 作者简介:刘耀宗(1974-),男,博士研究生,主研方向:数据流挖掘;张 宏,教授、博士生导师;孟 锦,博士研究生;韩法旺,讲师、硕士

Outliers Detection in Data Stream Based on Wavelet Density Estimation

LIU Yao-zong 1,2, ZHANG Hong 1, MENG Jin 1,2, HAN Fa-wang 2   

  1. (1. School of Computer, Nanjing University of Science and Technology, Nanjing 210094, China; 2. Department of Information, Nanjing Forest Police College, Nanjing 210046, China)
  • Received:2012-03-16 Revised:2012-05-31 Online:2013-02-15 Published:2013-02-13

摘要: 为能及时发现数据流上的局部离群点,分析数据流已有的离群点挖掘算法,提出基于小波密度估计的离群点检测算法。利用小波密度估计多尺度和多粒度的特点,通过小波概率阈值判断数据流中当前滑动窗口内的数据点是否为离群点,并对数据流中离群点检测过程进行讨论。仿真结果表明,与核密度估计算法相比,该算法的检测效率与精度较高。

关键词: 数据流, 局部离群点, 离群点检测, 核密度估计, 小波密度估计

Abstract: In order to find local outliers in data stream, this paper analyzes traditional outliers detection algorithms, and intro- duces an outlier detection algorithm based on Wavelet Density Estimation(WDE). It uses a multi-scale and multi-granularity characteristics of WDE using the wavelet probability threshold to judge the data stream within the current sliding window data points as outliers, and discusses outlier detection in data stream process. Simulation results show that this algorithm has higher detection efficiency and accuracy in the data stream than Kernel Density Estimation(KDE) algorithm.

Key words: data stream, local outlier, outlier detection, Kernel Density Estimation(KDE), Wavelet Density Estimation(WDE)

中图分类号: