Geographic Data Mining and Knowledge Discovery

Abstract

The strength of a geographic information system (GIS) is in providing a rich data \ninfrastructure for combining disparate data in meaningful ways, by using a spatial \narrangement (e.g., proximity). As a toolbox, a GIS allows planners to perform spatial \nanalysis \nusing geo-processing functions, such as map overlay, connectivity measurements, \nor thematic map coloring. Although this makes the geographic visualization \nof individual variables effective, complex multi-variate dependencies \nare easily overlooked. \nThe required step to take GIS beyond a tool for automating cartography is to \nincorporate the ability of analyzing and condensing \na large number of geo-referenced \nvariables into a single forecast or score. This is where spatial data mining promises \ngreat potential benefits and the reason why there is such a hand-in-glove fit between \nGIS and data mining facilities. INGENS 2.0 is a prototype GIS which resorts to \nemerging spatial data mining technology to support geographers, geologists, and \ntown planners in discovering (descriptive and predictive) patterns, which are never \nexplicitly represented in topographic maps or in a GIS-model and are useful in the \ntask of topographic map interpretation. In spatial data mining, spatial \ndimension adds \na substantial complexity to the data mining task. First, spatial objects are characterized \nby a geometrical representation and relative positioning with respect to a \nreference system, which implicitly define spatial properties. Modeling these implicit \nspatial properties (attributes and relations) \nin order to associate them with clear \nsemantics and a set of eficient procedures for their computation is the first challenge \nto be met when facing a spatial data mining problem. Second, spatial phenomena \nare characterized by autocorrelation, i.e., observations of spatially distributed random \nvariables are not location-independent. Third, spatial objects can be considered \nat different \nlevels of abstraction (or granularity). Spatial data mining facilities in \nINGENS deal with these challenges in both inducing classification rules and discovering \nassociation rules from spatial data. The spatial data mining process \nis aimed \nat a user who controls the parameters of the process by means of a query written in \nSDMOQL, a spatial data mining query language that permits the specification of the \ntask-relevant data, the kind of knowledge to be mined, the background knowledge \nand the hierarchies and the interestingness \nmeasures. Some constraints on the query \nlanguage are identified by the particular mining task. An application to a real repository \nof topographic maps is briefly illustrated.