Edwin M. Knorr, Raymond T. Ng
We study the spatial data mining problem of how to extract a special type of proximity relationship - namely that of distinguishing two clusters of points based on the types of their neighbouring features. The points in the clusters may represent houses on a map, and the features may represent spatial entities such as schools, parks, golf courses, etc. Classes of features are organized into concept hierarchies. We develop algorithm GenDis which uses concept generalization to identify the distinguishing features or concepts which serve as discriminators. Furthermore, we study the issue of which discriminators are "better" than others by introducing the notion of maximal discriminators, and by using a ranking system to quantitatively weigh maximal discriminators from different concept hierarchies.