Density-Connected Sets and their Application for Trend Detection in Spatial Databases

Several clustering algorithms have been proposed for class identification in spatial databases such as earth observation databases. The effectivity of the well-known algorithms such as DBSCAN, however, is somewhat limited because they do not fully exploit the richness of the different types of data contained in a spatial database. In this paper, we introduce the concept of density-connected sets and present a significantly generalized version of DBSCAN. The major properties of this algorithm are as follows: (1) any symmetric predicate can be used to define the neighborhood of an object allowing a natural definition in the case of spatially extended objects such as polygons, and (2) the cardinality function for a set of neighboring objects may take into account the non-spatial attributes of the objects as a means of assigning application specific weights. Density-connected sets can be used as a basis to discover trends in a spatial database. We define trends in spatial databases and show how to apply the generalized DBSCAN algorithm for the task of discovering such knowledge. To demonstrate the practical impact of our approach, we performed experiments on a geographical information system on Bavaria which is representative for a broad class of spatial databases.