Applying Relational Dependency Discovery Framework to Geo-spatial Data Mining

Identifying dependencies and relationships amongst data collected from disparate sources is a difficult problem. When the data to be analyzed are geo-spatial data, the problem becomes harder because meaningful data dependencies may not be derived until the right granularity of geo-spatial objects is used to organize the data. We propose a computational framework in which dependencies between geo-spatial referencing variables are automatically examined through trial and error. We present the computation model which uses a set of heuristic rules to derive meaningful dependencies. This model assumes the dependency derivation engine is to be used repeatedly to find the right granularity of the geo-spatial objects with which the discovered dependencies may exhibit most "interesting" relationships. We demonstrate how the model works with an example data set. We also illustrate that the uncovered dependencies can be depicted over a GIS system for easier understanding of the dependencies by the end users.