基于分布模型的层次聚类算法

提出了一种新的层次聚类算法,先对数据集进行采样,以采样点为中心吸收邻域内的数据点形成子簇,再根据子簇是否相交实现层次聚类。在层次聚类过程中,重新定义了簇与簇之间的距离度量,并以此为基础建立堆结构。利用估计数据点总体分布的思想,证明该算法将逼近最优解。实验结果表明,算法的聚类效果大大优于现有的聚类算法。

Hierarchical Clustering Algorithm Based on Distribution Model

A novel agglomerative method is proposed. This algorithm consists of three steps, first samples the dataset, then form the subcluster by absorbing the points in the ?neighborhoods of sample points, at last final clusters are constructed by combining the subclusters. The distance measure of two clusters is redefined. Based on this concept, heap structure is constructed. Formally a theoretical explanation of the algorithm is given using the method approaching the actual distribution. Experimental results show the quality of ADA is much better than very many well-known algorithm CURE.