一种多因素融合的高效离群点检测方法  

An efficient outlier detection method based on multi-factor fusion

作　　者：杨志强 冯山 尹伊 吴慧佳 Zhiqiang YANG;Shan FENG;Yi YIN;Huijia WU(College of Mathematics and Science,Sichuan Normal University,Chengdu 610068,Sichuan,China)

机构地区：[1]四川师范大学数学科学学院,四川成都610068

出　　处：《山东大学学报（理学版）》2024年第8期118-126,共9页Journal of Shandong University(Natural Science)

基　　金：国家自然科学基金资助项目(61673285,62306196);四川省自然科学基金资助项目(24NSFSC1566,24NSFSC1487)

摘　　要：基于邻域粗糙集的对象邻域相对比和对象重要度等粒化特征,提出了改进的基于邻域粗糙熵的多因素融合的离群点检测(neighborhood rough entropy-based outlier,NREOD)算法。在加利福尼亚大学尔湾分校(University of CaliforniaIrvine,UCI)数据库的标准数据集上的对比实验表明,NREOD算法在不同类型的数据集的离群检测的误判率更低,并且有更好的适应性和有效性。此算法为混合型属性数据集的离群检测研究与应用提供了一条新的有效途径。Based on granularity characteristics of relative ratio of object neighborhood and importance of objects in neighborhood rough sets,an improved outlier detection method(neighborhood rough entropy-based outlier,NREOD)based on neighborhood rough entropy and multi-factor fusion is proposed.Comparison experiments on standard data sets in University of CaliforniaIrvine(UCI)databases show that the NREOD algorithm has a lower false positive rate for outlier detection in different types of data sets,and has better adaptability and effectiveness.This algorithm provides a new effective way for the research and application of outlier detection in mixed attribute data sets.

关 键 词：数据挖掘 离群点检测 邻域粗糙集 邻域粗糙熵 多因素融合 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]