A New Hierarchical Clustering Algorithm Study And Application

Clustering analysis is a major field in data mining, which is an important method of data partition, especially is widely used in the fields of data mining, statistics, machine study, space database technology, biology and marketing analysis etc. Clustering algorithms includes partitioning, hierarchical, density-based, grid-based, model-based algorithm.This article focused on the analysis and research of hierarchical clustering algorithms, and given the merits and demerits of it. After the analysis for the performance of Chameleon algorithm, A Chameleon algorithm based on dynamic nearest neighbors selection model is presented.Community structure is an important characteristic in complex networks. Seeking and analyzing communities is an invaluable tool of understanding the structure of networks. In recent years, a lot of_algorithms have been proposed to detect the community structure in networks. In this dissertation, we study the community structure by hierarchical clustering algorithms.The main work of this thesis is as follows:(1)This thesis analyzes hierarchical clustering algorithms comprehensively, it also introduces the BIRCH algorithm, CURE algorithm, ROCK algorithm and Chameleon algorithm, the advantages and disadvantages of each algorithm are compared.(2)After the analysis for the performance of Chameleon algorithm, a new algorithm named DNMC is presented, which considers the backtracking mechanism making DNMC benefit to the decomposition after the merger. Experimental results on databases Wine and Iris demonstrate that DNNC outperforms M-Chameleon based on the evaluation metrics of f_a. Following the calculation of disparity of each attribute, it is found that some attributes have little effect on the results of clustering. Therefore, the complexity of the time can be improved if those attributes are neglected.(3)This thesis briefly introduces the algorithms of detecting community structure in complex networks, presents detailed simulation experiments. Experimental results indicate that it is feasible and effective to apply the improved algorithm to solving the problem of object grouping.