粗糙集中的距离度量与离群点检测-控制与决策.pdfVIP

第28 卷 第2 期 控 制 与 决 策 2013 年 2 月 Vol. 28 No. 2 Control and Decision Feb. 2013 文章编号: 1001-0920 (2013) 02-0188-05 粗糙集中的距离度量与离群点检测 1 2 2 江 峰 , 眭跃飞 , 曹存根 (1. 青岛科技大学信息科学技术学院，山东青岛266061 ；2. 中国科学院计算技术研究所，北京100080) 摘 要: 针对传统的基于距离的离群点检测方法不能有效地处理具有离散型属性数据集的问题, 将基于距离的离群 点检测方法引入粗糙集理论, 利用粗糙集解决离散型属性的处理问题. 首先, 在粗糙集的框架中提出3 种面向离散型 属性的距离度量; 然后, 针对这3 种距离度量分别设计出相应的离群点检测算法, 用于从包含离散型属性的数据集中 检测离群点; 最后, 通过在2 个包含离散型属性的UCI 数据集上的实验, 验证了这些算法的可行性和有效性. 关键词: 粗糙集理论；离群点检测；数据挖掘；距离度量；离散型属性 中图分类号: TP311 文献标志码: A Distance metrics and outlier detection in rough sets 1 2 2 JIANG Feng , SUI Yue-fei , CAO Cun-gen (1. College of Information Science and Technology，Qingdao University of Science and Technology，Qingdao 266061, China ；2. Institute of Computing Technology，Chinese Academy of Sciences，Beijing 100080 ，China ．Correspondent: JIANG Feng ，E-mail ：jiangkong@163.net) Abstract ：：：The traditional distance-based outlier detection method can not effectively deal with the data sets containing discrete attributes. Therefore, the distance-based outlier detection method to rough sets is introduced, and the advantage of rough sets is taken to solve the problem of dealing with discrete attributes. First, three distance metrics for discrete attributes within the framework of rough sets are proposed. Second, for each of these distance metrics, a corresponding outlier detection algorithm is designed, to detect outliers from data sets containing discrete attributes. Finally, the feasibility and effectiveness of these algorithms are demonstrated on two UCI data sets containing discrete attributes. Key wo