一种基于近邻分布的空间聚类方法

A Spatial Clustering Algorithm Based on Neighbor Distribution

摘要: 空间聚类是分析空间数据的一种重要工具，空间数据库中的聚类形状可能会是任意形状的，比如球状的、持续的、延伸的等，因此，能否发现任意形状簇成为衡量空间聚类算法质量的一个重要标准。本文中我们提出一种基于最近邻距离分布的空间聚类方法，这个算法是基于这样一个假设，假设在数据空间的某个特定部分，一个聚类内部的点是均匀分布的。实验分析表明，该方法不仅能发现任意形状的簇，而且对于大型空间数据库是高效有用的。

Abstract: The spatial clustering is an important tool for analyzing spatial data. Clusters in the spatial database may be of arbitrary shape e.g. spherical, drawn-out, linear, elongated etc. Therefore, to discover clusters of arbitrary shape be an important quality standard for spatial clustering algorithm. In this paper, we present a spatial clustering algorithm based on the nearest distance distribution, the algorithm is based on a reasonable assumption that in the particular part of data space, points within same cluster are uniformly distributed. Experimental analysis shows that the method can discover clusters of arbitrary shape and it is efficient and useful for large spatial database.