聚类分析的一种改进方法

An Improvement on Cluster Analysis

为了削减由于指标变量不同量纲、不同权重及其相关性对聚类效果造成的影响,把权重引入马氏距离计算公式,定义了加权马氏距离,并证明了其在理论上的有效性.最后,采用加权马氏距离对一个具体算例进行聚类分析,得到了较为理想的、更符合实际的分类结果,表明了该定义在应用上的有效性.

To diminish the influence on clustering effect resulting from different measurements, different weights and correlation of indicator variables. weights are brought in Mahalanobis distance in this paper. Then weighted Mahalanobis distance is defined and proved valid in theory. At last, a clustering analysis is then conducted for a specific case based on weighted Mahalanobis distance, which generates more reasonable and more genuine classified result, indicates the validity of weighted Mahalanobis distance in practice.