Visualization and Database Support for Geographic Meta-Mining

Introduction Geographic data mining can be defined as a set of exploratory computational and statistical approaches for analyzing very large spatial and spatiotemporal data sets. Data mining techniques are often grouped into categories that include clustering, categorization, summarization, rule-mining, and feature extraction. All of these types of techniques are generally oriented towards identifying spatial or spatio-temporal patterns in geographic observations or measurements. The identification of these patterns is intended to spur hypothesis generation as to the geographic processes from which the patterns are generated. Data mining can be considered one step in the larger process of geographic knowledge discovery (Fayyad et al. 1996). This process is both interactive and iterative and includes steps such as data selection, data cleaning, and the interpretation of data mining results. While there are a variety of academic and commercial data mining software available, there are few comprehensive knowledge discovery software environments. The process of knowledge discovery is supported wholly by the analyst, who is responsible for keeping tract of the results of multiple data mining 'runs,' for instance descriptions of rules or extracted features. These results are ideally elements of knowledge – summaries of patterns embedded within the observational data. However, these results can also be considered a form of data themselves that are often in need of further analysis to yield useful interpretation. In many cases, the data sets resulting from data mining are large and complex, yet there are few computational techniques for managing these data mining results to fully support the knowledge discovery process. The analysis of the results of data mining is called meta-mining (Abraham and Roddick 1999). I argue here that geographic knowledge discovery software demands support not only for data mining but also for the ability to visually and algorithmically meta-mine the results of data mining in an interactive and iterative manner. Computational support for meta-mining is dependent on the visualization and database representation of the rules, clusters, and features that are the results of data mining. Thus, geographic knowledge discovery software environments must incorporate visualization and semantic database modeling strategies for the representation of, and interaction with, these knowledge elements.