基于自适应权重的面板数据聚类方法  

Clustering method for panel data based on adaptive weighting

作　　者：李因果[1] 戴翼[2] 何晓群[2] 

机构地区：[1]江苏师范大学经济学院,徐州221116 [2]中国人民大学应用统计科学研究中心,北京100872

出　　处：《系统工程理论与实践》2013年第2期388-395,共8页Systems Engineering-Theory & Practice

基　　金：中国人民大学研究生科学研究基金(22396180)

摘　　要：基于二维信息的传统聚类方法并不适用于处理面板数据,在考察面板数据多重信息特征的基础上,基于面板数据的"绝对指标","增量指标"及"波动指标",重构了面板数据相似性测度的距离函数和Ward聚类算法,提出了面板数据自适应权重聚类方法.所提供的算法既可退化为传统的绝对量距离聚类方法,亦可对面板数据的未来所属类别进行聚类预测.最后,实例显示此方法兼具有效性和灵活性.The traditional clustering methods based on the two-dimension information are not suitable for dealing with the panel data processing, in the consideration of the multiple information characteris- tics of panel data and according to the ＂absolute index＂, ＂incremental index＂ and ＂volatility index＂ of the panel data, this paper reconstructed the similarity measure distance function and Ward clustering algorithm, proposed an adaptive weighting clustering method for panel data. The provided algorithm not only can degenerate into the traditional ＂absolute distance＂ clustering algorithm, but also can conduct clustering prediction for the future category of panel data. Finally, examples show that this method is both effectiveness and flexibility.

关 键 词：面板数据 增速指标 波动指标 自适应权重 聚类预测 

分 类 号：F222.3[经济管理—国民经济]