计算机工程

• 人工智能及识别技术 • 上一篇    下一篇

基于蚁群算法的异常数据检测方法

蔡美,刘波   

  1. 基于蚁群算法的异常数据检测方法
  • 出版日期:2016-08-15 发布日期:2016-08-15
  • 作者简介:蔡美(1990-),女,硕士研究生,主研方向为群体智能、数据挖掘;刘波(通讯作者),教授。
  • 基金项目:

    国家自然科学基金资助项目(U1431227);广东省科技计划基金资助项目(2013B010401017);广州市科技计划基金资助项目(201504290939226)。

Abnormal Data Detection Method Based on Ant Colony Algorithm

CAI Mei,LIU Bo   

  1. Abnormal Data Detection Method Based on Ant Colony Algorithm
  • Online:2016-08-15 Published:2016-08-15

摘要:

由于传统基于Omeasure度量的异常数据检测方法在对异常数据进行检测时需要对路径进行全搜索,并且在数据量较少的情况下会产生误判,因此其在检测效率和查准率上具有明显的缺陷。为此,根据蚁群算法的正反馈性质,提出一种将蚁群算法和属性相关分析相结合的属性异常点检测方法。将蚁群收敛到的路径作为异常路径,计算异常路径上各个节点Omeasure值,并根据Omeasure值确定数据异常点。实验结果表明,该方法在查全率、查准率和效率上均优于传统的基于Omeasure度量的异常数据检测方法。

关键词: 异常检测, 属性, 异常数据, 蚁群算法, 路径选择

Abstract:

Since the traditional outlier detection technology based on Omeasure needs to search all paths while detecting abnormal data and it is easy to make misjudgments under the scenario of less amount of data.Hence,it has obvious defects on the efficiency and precision ratio.According to the positive feedback feature of ant colony algorithm,a method which combines ant colony algorithm and attribute correlation analysis is put forward for attribute outlier detection.The method chooses the converged paths of the ant colony as the exception paths,then computes Omeasure value of each node on those paths,and identifies the outlier based on the Omeasure values.Experimental results show that this method performs better in recall,precision and efficiency than traditional outlier detection technology based on Omeasure.

Key words: anomaly detection, attribute, abnormal data, ant colony algorithm, path selection

中图分类号: