An Integrated Environment for High-dimensional Geographic Data Mining

Introduction Geographic data are often very large in volume and “characterized by a high number of attributes or dimensions” [1]. There are urgent needs to develop effective and yet efficient approaches for analyzing such voluminous and high-dimensional data to address complex geographic problems [1, 2, 3, 4], e.g., detecting unknown multivariate patterns or relationships between socioeconomic, demographic, environmental factors and the incidence of various cancers. This paper introduces an integrated geographic data mining environment, which couples a suite of visualization and computational methods to explore multivariate patterns in large and high-dimensional geographic datasets. The integrated geographic data mining environment involves four major groups of components: (1) interactive feature selection components to identify interesting subsets of variables for further analysis[5]; (2) self-organizing map (SOM) [6] components to cluster data objects with only the variables selected above; (3) a high-dimensional visualization component—Parallel Coordinate Plot (PCP) [7]—to explore and present multivariate patterns or relationships; and (4) a geographic map component to visualize the spatial distribution of patterns discovered above. With interactive manipulation of these integrated components, the user can iteratively locate, interpret, and refine patterns.