基于集聚度增量的空间聚类算法

A Fast Algorithm of Spatial Clustering Based on Agglomeration

该文借鉴了复杂网络领域的模块度概念,构造了空间点集的集聚度函数。基于集聚度的增量值,提出一个快速的空间聚类算法。实验证明,该值同点集的类间均方差(SSB)与类内均方差(SSE)的比值(SSB/SSE)有相同的结论,可以评价不同的点集在空间分布上的集聚程度(即群簇结构是否明显),同时该算法可以在不预先设定聚类个数的情况下快速有效地得到聚类结果。

In the field of spatial analysis, clustering is always under the spotlight. Many methods for cluster detection have been well studied. Among them, there exists a key issue which many researchers are concentrated on. That is how to get the optimal clustering results when we don＇t know the number of clusters beforehand. In the field of complex networks, modularity is used to measure the clustering of links. Based on the modularity, a definition of agglomeration was proposed to measure the spatial clustering structure. Then, a fast algorithm was put forward for grouping space points based on the increment of agglomeration. The experiments show that the value of agglomeration can evaluate the clustering structure between different datasets, and the result is similar to the SSB/SSE（SSB is the variance between clusters, and SSE is the variance within the clusters）. In additional, the algorithm runs quickly and effectively.