Abstract: A spatial co-location pattern is a set of spatial features whose instances frequently appear together in a geographic space. The traditional co-location pattern mining usually determines the proximity relationship of two instances by giving a single proximity threshold. Using a single proximity threshold may result in the loss of the proximity relationships, and does not consider the influence of different distances on the proximity relationships. In addition, the efficiency of the algorithm is sensitive to the prevalence thresholds. Because frequently co-located features have a high degree of similarity, they can be clustered together by using clustering algorithms. Moreover, the proximity relationship and the co-located relationship of features are fuzzy concepts. Thus, based on the definitions of the fuzzy proximity relationship and the feature proximity measure, this paper studies fuzzy mining technique in spatial co-location pattern mining and proposes a method of mining the spatial co-location patterns by combining the fuzzy set theory and the clustering algorithm. The practicability, efficiency and robustness of the proposed method are proved by extensive experiments.

Key words: spatial data mining, spatial co-location pattern, fuzzy proximity relationship, fuzzy clustering

摘要： 空间并置（co-location）模式是指其特征的实例在地理空间中频繁并置出现的一组空间特征的集合。传统co-location模式挖掘通常由用户给定一个邻近阈值来确定实例的邻近关系，使用单一的邻近阈值来判定两个空间实例的邻近性可能会造成邻近关系的缺失，也没有考虑距离大小的不同对邻近关系的影响。同时，传统方法主要利用频繁性阈值来衡量模式的频繁性，存在着算法效率对频繁性阈值较为敏感的问题。由于频繁并置的特征间具有较高的邻近度，因此利用聚类算法可以将其聚集在一起，加之邻近以及特征间的并置都是模糊的概念，因此将模糊集理论与聚类算法相结合，研究了空间co-location模式挖掘中的模糊挖掘技术，在定义模糊邻近关系的基础上，定义了度量特征之间邻近度的函数，基于特征邻近度利用模糊聚类算法挖掘co-location模式，最后通过广泛的实验验证了提出方法的实用性、高效性及鲁棒性。

关键词: 空间数据挖掘, 空间co-location模式, 模糊邻近, 模糊聚类