一种基于多维空间超球体的快速聚类算法

提出了一种基于多维空间超球体的快速聚类算法。这种算法结合密度聚类和层次聚类两种思想。首先利用密度聚类方法将小范围内的数据对象聚合成超球体,然后再按照层次聚类中的凝聚思想,根据超球体之间的位置关系产生最终聚类。实验表明,该算法对于数值型数据集不仅在效率、噪声敏感性等方面均有较好的表现,同时还可以通过诸如“双环测试”等带有刁难性的特殊测试集。以往,常常简单的以距离来刻画的数据间“相似性”,而所提出的快速聚类算法则改由超球体之间连接的紧密程度来描述这种性质。实验表明,这种修改使得算法的性能得到了很好的改善。

Fast Clustering Algorithm Based on Hypersphere of Multidimensional Space

A fast clustering algorithm based on hypersphere of the multidimensional space is presented.The clustering method combines the density and the hierarchical clustering methods.Firstly,it makes the hyperspheres by the data in a small area.Then it works according to the agglomerative hierarchical clustering method. The final cluster related to the relationship of different hyperspheres is produced.Experiments show that the algorithm has done well in the efficiency and the noises sensitive on the numerical data set.It can also pass some difficult tests such as "double circle".The similar data are usually depicted as the distance,however,in the algorithm the compactness is used instead of the distance.The experiment shows that the performances of the algorithm are improved.