基于谱聚类的多维数据集异常子群挖掘方法  

Method for Mining Abnormal Subgroups of Multidimensional Data Sets Based on Spectral Clustering

作　　者：康耀龙 冯丽露 张景安[3] KANG Yao-long;FENG Li-lu;ZHANG Jing-an(School of Computer and Network Engineering,Shanxi Datong University,Datong Shanxi 037009,China;Shanxi Datong University,Datong Shanxi 037009,China;Computer Network Center,Shanxi Datong University,Datong Shanxi 037009,China)

机构地区：[1]山西大同大学计算机与网络工程学院,山西大同037009 [2]山西大同大学,山西大同037009 [3]山西大同大学计算机网络中心,山西大同037009

出　　处：《计算机仿真》2023年第7期477-480,523,共5页Computer Simulation

基　　金：教育部产学合作协同育人项目(201902126005);大同市平台基地计划项目(2020196);山西大同大学基础研究项目(2022K9,2022K16);山西大同大学教学改革创新项目(XJG2022201,XJG2022202)。

摘　　要：针对传统方法存在的异常子群挖掘结果准确性不高,挖掘效果不佳的问题,提出基于谱聚类的多维数据集异常子群挖掘方法。通过多维数据集预处理判断显著子群,依据属性值构建同阶子群,获取数据集中存在的部分候选子群;采用基于L1范数的约束谱聚类算法划分候选子群后,利用约束矩阵、二分类以及整合,完成候选子群的多分类、处理正约束点以及重现顶点的度和边,形成约化图并完成候选子群的挖掘,即实现异常子群挖掘。测试结果表明,上述方法的挖掘准确率和标准化互信息值较高,可完成指定异常子群的多维深度挖掘,且挖掘效果良好。In traditional methods,the results of abnormal subgroup mining are often not accurate and the mining effect is not ideal.In this article,a method for mining abnormal subgroups in multidimensional data sets based on spectral clustering was put forward.At first,we judged the significant subgroups through the preprocessing for multi-dimensional data sets.According to attribute values,we constructed the subgroups of the same order and thus ob-tained some candidate subgroups in data sets.After the candidate subgroups were divided by the constraint spectral clustering algorithm based on L1 norm,we used the constraint matrix,binary classification and integration to com-plete the multi-classification of candidate subgroups,the processing of positive constraint points and reappearance of the degree and edge of a vertex.Afterwards,the reduced graph was formed,and the mining of candidate subgroups was finished.Finally,we achieved the mining for abnormal subgroups.The test results prove that the proposed method has high mining accuracy and high standardized mutual information value,so it can achieve the multi-dimen-sional deep mining of specified abnormal subgroups,with good effect.

关 键 词：谱聚类 异常子群挖掘 显著子群 候选子群 约化图 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]