划分法在分类效果上的一些理论探讨

划分法是最基本的聚类方法，常用的有k-means、K-medoids和它们的改进方法。当结果簇是密集的，且簇与簇之间区别明显时，它们的分类效果很好。但对非球状的簇或者大小相差大的簇却不能很好地分开。本文从方差的角度分析了划分法对于不同的数据类在分类效果上存在差异的原因。

A Theoretical Exploration of Classifying Effect of Partitioning Method

Partitioning methods are a major clustering method. The commonly used methods of it include k - means, k - medoids and their improvements. The work well when the clusters are compact clouds that are rather well separted from one another. However,they are not suitable for discovering clusters with non - globular shapes or clusters of very different size. This article analyzes the causes for the existent differences of effects of portioning on classification from the perspective of variation.