Discovering Knowledge from High-Dimensional Geographic Data: Integrating Visual and Computational Approaches

It has been widely recognized that spatial data analysis capabilities have not kept up with the need for analyzing the increasingly large volumes of geographic data of various themes that are currently being collected and archived (Openshaw 1991; Miller and Han 2001; Shekhar, Vatsaval et al. 2002; Guo 2003; Guo, Peuquet et al. 2003; Muntz, Barclay et al. 2003). On one hand, such a wealth of data holds great opportunities for geographers, environmental scientists, public health researchers, and others to address urgent and sophisticated geographic problems, e.g., global change, epidemics such as SARS, etc. On the other hand, existing data analysis methods fall short for the extraction of meaningful patterns from datasets of such unprecedentedly large size (in terms of the number of observations) and high dimensionality (in terms of the number of variables). Data mining and knowledge discovery refers to the overall process of discovering useful knowledge from data, which generally involves data selection, data pre-processing, data transformation, incorporation of appropriate prior knowledge, data mining, and proper interpretation of the results (Fayyad, Piatetsky-Shapiro et al. 1996). While data mining and KDD research has been widely conducted in areas of business, bioinformatics, text mining, etc., it is still at a very early stage in geographic domains. Geography is an integrative discipline and geographic data under analysis often span across multiple domains. The complexity of spatial data and geographic problems, together with intrinsic spatial relationships, constitute an enormous challenge to conventional data mining methods and call for both theoretical research and development of new techniques to assist in deriving information from large and heterogeneous spatial datasets (Han and Kamber 2001; Miller and Han 2001; Gahegan and Brodaric 2002).