Mining Model Trees from Spatial Data

Mining regression models from spatial data is a fundamental task in Spatial Data Mining. We propose a method, namely Mrs-SMOTI, that takes advantage from a tight-integration with spatial databases and mines regression models in form of trees in order to partition the sample space. The method is characterized by three aspects. First, it is able to capture both spatially global and local effects of explanatory attributes. Second, explanatory attributes that influence the response attribute do not necessarily come from a single layer. Third, the consideration that geometrical representation and relative positioning of spatial objects with respect to a reference system implicitly define both spatial relationships and properties. An application to real-world spatial data is reported.