基于多指标面板数据的改进的聚类方法及应用  

An improved clustering method of the multivariable panel data and its application

作　　者：高雪[1] 谢仪[1] 侯红卫[1] 

机构地区：[1]太原理工大学数学学院,山西太原030024

出　　处：《浙江工业大学学报》2014年第4期468-472,共5页Journal of Zhejiang University of Technology

基　　金：国家自然科学天元基金资助项目(11326126)

摘　　要：针对数据挖掘中多指标面板数据的聚类分析问题,构建了一种新的对象间距离的定义,并基于传统的k均值聚类方法,将时间维度分割,对每相邻时间段的对象进行聚类.将单次聚类结果汇总形成聚类结果矩阵,根据汇总的结果矩阵计算对象归于某一类的隶属权值,从而确定最终的聚类结果.这样同时考虑对象在空间上和时间上的发展趋势的聚类方法将得到更加全面客观的聚类结果.最后将本聚类方法运用于金融保险行业上市公司财务数据,进行实证分析,指出该方法的有效性.For the clustering analysis of the multivariable panel data in data mining,a new definition is constructed for the distance between samples.Then the samples at every adjacent time period are clustered after split the time series,based on the traditional k-means ways of clustering.Thus a matrix of clustering result is formed by merging with single clustering results.Meanwhile the calculation of membership based on the former matrix is presented.Comparing with the clustering method which only considers the spacial development trend of the samples,this improved method can obtain a much better and objective result.As an example,the algorithm is used to deal with the analysis of financial data of listed corporation in the financial and insurance industry in our country to put forward the validity and applicability of the method in this paper.

关 键 词：数据挖掘 多指标面板数据 综合距离 隶属权值 聚类分析 

分 类 号：O212.4[理学—概率论与数理统计]