一种面向混合属性数据的邻域粒离群点检测

Outlier Detection Based on Neighborhood Granule for Mixed Attribute Data

针对粗糙集方法不能有效处理数值和混合型数据的问题,本文以邻域粗糙集中粒计算为基础,提出了一种基于邻域粒的离群点检测方法.首先,给出了邻域粒之间距离的定义.其次,定义粒离群程度和基于邻域粒的离群因子来分别表征邻域粒和对象的离群程度,设计并实现了基于邻域粒的离群点检测(NGOD)算法.最后,利用实际数据集对NGOD的有效性进行了评估.实验结果表明,所提检测方法对分类、数值和混合属性数据是有效的.

Aiming at the problem that the rough set method cannot effectively process the numerical and mixed data,based on granular computing in the neighborhood rough set,this paper proposes a neighborhood granule based outlier detection method.Firstly,the definition of the distance between neighborhood granules is given.Secondly,the granule outlier degree and neighborhood granule-based outlier factor are defined to characterize the abnormality of neighborhood granules and objects respectively.The Neighborhood Granule-based Outlier Detection(NGOD)algorithm is designed and implemented.Finally,the effectiveness of the NGOD was evaluated using actual data sets.The experimental results show that the proposed detection method is effective for categorical,numerical and mixed attribute data.