具有层次结构的K-means聚类算法研究

Research on An Improved Hierarchical Clustering Algorithm of K-means

提出一种改进的基于层次微聚类的K-means聚类算法,并重新构造准则函数S(k).通过层次聚类生成一颗层次聚类树,根据微聚类思想在该聚类树上动态更新中心点,利用改进的准则函数S(k)选择合理聚类个数K和对应中心点,使得聚类结果达到全局最优.标准数据集上的实验结果表明,与传统K-means聚类算法相比,改进后K-means聚类算法的聚类效果和聚类精度都有较大提高.

An improved K-means clustering algorithm based on hierarchical micro-clustering（HMKC）was proposed,combining the hierarchical structure of space,and the criterion function S（k）was reconstructed.Firstly,a hierarchical K-means clustering tree was produced by using hierarchical clustering algorithm,then the center points were updated dynamically on the tree structure according to the micro-clustering method.Finally,the improved criterion function S（k）was used to find rational clustering number K and corresponding core point to ensure the clustering result has reached optimal globally.Experimental results on standard datasets demonstrate that the effect and accuracy of clustering results can be improved significantly with the HMKC algorithm comparing with the traditional k-means algorithms.