一种改进的Chameleon聚类算法

在聚类分析技术中Chameleon算法在发现高质量任意形状簇方面具有优势。但是该算法需要用户人为给出K-最近邻的K值以及停止合并子簇的位置等参数，在没有先验知识的情况下确定这些参数难度较大。此外，Chameleon算法的第一步中需要用到图划分技术，这是一个NP-难问题，因此大大增加了算法的复杂度。本文引入模块度的概念，提出了一种改进的Chameleon层次聚类算法（I-Chameleon算法）很好地解决了以上诸多问题，能够自动地发现任意形状簇。

Improved Chameleon Algorithm for Cluster Analysis

The Chameleon algorithm is good at discovering arbitrary shape Clusters with high quality among cluster analysis technologies. But this algorithm needs user to give the K-most close neighbor's K value as well as the position to stop Combing sub-Clusters artificially. It is difficult to determinate these parameters without prior knowledge. Besides that, the Chameleon algorithm needs to use graph partitioning technology which is a NP-hard problem in the first step therefore increases the Computation order of complexity greatly. This paper proposes one kind of improved hierarchical clustering algorithm (I-Chameleon algorithm) which could solve the above many problems very well and find arbitrary shape clusters automatically by introducing the concept of modularity.