Spatial Association Rule Discovery with Rule Classification and Variable Sensitivity

Geographic information processing and spatial data analysis activities have increased dramatically in recent years, and data mining techniques have the potential to automate some of these processing tasks. Spatial data mining can be used to identify relationships between features in a data set and to discover and validate association rules. An association rule discovery technique is proposed which can work with data from multiple sources, can handle both discrete features and continuous variables, and can account for spatial neighborhood effects. The Phase I work demonstrated the feasibility of the concept, using a common data format along with predicate and rule formulations that include feature, variable, and spatial representations. This work resulted in the discovery of spatial association rules, and also identified two critical areas for further research: limiting the rules that are tested and identifying the significant variables in a rule. The Phase II work will develop a classification system for rule formulation, improve the spatial representation, perform sensitivity studies to identify significant variables, extend the data management, develop a user interface, and integrate with a GIS. The system will provide association rules to geographic analysts for use in validation, error checking, or prediction.