文摘
Much effort has been taken in knowledge discovery for geographic classification and category extraction. Previous studies have been limited to extracting categories with definite boundaries and uniform memberships within class boundaries. This dissertation presents a framework for representing and extracting knowledge for categorizing geographic entities with fuzzy boundaries. The framework is based on prototype theory and consists of three components: (1) a knowledge representation scheme that capture the prototype effects of geographic categories, (2) a knowledge discovery procedure that enables the extraction of knowledge in such representation, and (3) an inference scheme that uses the extracted knowledge to classify geographic entities while maintaining the prototypical properties of the resulting categories. Three data mining approaches are also presented in this dissertation to populate the framework. A case study on knowledge discovery for fuzzy soil mapping demonstrates the implementation of the framework. The case study shows that knowledge for classifying geographic entities with fuzzy boundaries is embedded in geographic data sources such as area-class maps and can be extracted under the framework. Such knowledge, represented in a way that is cognitively sound, captures not only the typical characteristics of the classes, but also within-class variations and transitions between classes.