一种基于部分样本类别判定的聚类分析方法  

A Method of Clustering Analysis Based on Category Judgement by Part Samples

作　　者：郭亚军[1] 王春震[1] 张发明[1] 邹家兴[1] 

机构地区：[1]东北大学工商管理学院,辽宁沈阳110004

出　　处：《东北大学学报（自然科学版）》2009年第7期1051-1054,共4页Journal of Northeastern University(Natural Science)

基　　金：国家自然科学基金资助项目(70472032)

摘　　要：针对传统聚类分析中,指标权重一般由专家直接给出,然后再在此基础上进行聚类分析的不足,提出了一种基于部分样本类别判定的聚类分析方法.首先对部分样本进行类别归属判定,然后利用类内聚类样本之间的距离应尽可能小的原理建立规划模型,通过"反推"的方式诱导出合理的权重信息,再据此进行样本聚类.该方法主要用于解决聚类样本较多,且聚类样本的指标权重难以显性确定情况下的聚类分析问题.最后给出的一个算例验证了所提方法的有效性.In view of that the conventional clustering analysis is based on such index weights that are generally given directly by the experts, a new method of clustering analysis is proposed on the basis of category judgement by part samples. Categorizing the part samples to judge them, a programming is developed on the principles that the distances between clustering samples in the same category should be shortened as possible. With the rational data of weights induced via counter-inference, the sample clustering is implemented. The method is mainly used to solve the problem of cluster analysis to which many samples are to be clustered and their index weights are difficult to determine explicitly. A numerical example is given to illustrate the effectiveness of the proposed method.

关 键 词：样本 类别判定 指标 权重 聚类 

分 类 号：N945[自然科学总论—系统科学] TP391[自动化与计算机技术—计算机应用技术]