• 中国计算机学会会刊
  • 中国科技核心期刊
  • 中文核心期刊

计算机工程与科学

• 论文 • 上一篇    下一篇

基于粒子群优化算法的属性异常检测

俞家宗,刘波   

  1. (暨南大学信息科学技术学院,广东 广州 510632)
  • 收稿日期:2015-10-14 修回日期:2016-03-17 出版日期:2017-06-25 发布日期:2017-06-25
  • 基金资助:

    国家自然科学基金(U1431227);广东省科技计划(2013B010401017);广州市科技计划(201604010037)

Attribute outlier detection based on particle
swarm optimization

YU Jia-zong,LIU Bo   

  1. (College of Information Science and Technology,Jinan University,Guangzhou 510632,China)
  • Received:2015-10-14 Revised:2016-03-17 Online:2017-06-25 Published:2017-06-25

摘要:

提出一种新的基于粒子群优化算法的属性异常检测算法。该算法利用粒子群优化算法简单、寻优速度快的优点检测属性异常,在粒子群寻找最优值的过程中发现可能是属性异常的数据,并采用O-measure适应度评估属性异常,算法的时间复杂度是多项式级的。与全搜索检测算法相比,大幅减少了搜索范围;同时,与完全随机算法相比,采用启发式搜索规则,提高了查全率及查准率。实验结果表明,粒子群检测算法不仅执行效率高,而且保持了较高的查全率与查准率。

 

关键词: 粒子优化群算法, 属性异常, 异常检测

Abstract:

We put forward a new attribute outlier detection algorithm based on particle swarm optimization, which applies its simplicity and rapid convergence to detect abnormal attributes. In the process of looking for the optimal value, abnormal attribute candidates are discovered, and they are assessed by the O-Measure fitness. The algorithm’s time complexity is polynomial. Compared with the full search algorithm, the proposed method can greatly reduce search scope; in comparison with the completely random algorithm, it uses heuristic search rules, thus the recall and precision are improved. Experimental results show that the particle swarm algorithm not only has high execution efficiency, but also maintains higher recall and precision rate. 
 

Key words: particle swarm optimization, attribute outlier, outlier detection