Abstract: Based on Chameleon algorithm, DPC algorithm and mudule density function, a hierarchical clustering algorithm based on density is proposed. In the first stage of Chameleon algorithm, DPC algorithm is used for data processing. In the second stage, when clusters are merged with the approximation function, module density that characterizes the similarity of data within a cluster is introduced. When module density achieves maximum,termination condition and final clustering result are obtained. An improved Chameleon algorithm is used to establish dynamic model, which can automatically determine the termination condition, and can identify many shape clusters, helping traditional Chameleon algorithm find the clustering end and be less sensitive to the initial parameter setting.

Key words: Chameleon algorithm, DPC algorithm, module density, robustness