Temporal association rule methodologies for geo-spatial decision support*.

摘要

This dissertation presents data mining algorithms that enable knowledge discovery in the framework of an intelligent, distributed Geo-spatial Decision Support System (GDSS). It provides an overview of the GDSS framework and uses the National Agricultural Decision Support System (NADSS) [17] to demonstrate the effectiveness of building knowledge discovery into a GDSS.; The data mining approaches that are developed, Representative Episodal Association Rules (REAR), and Minimal Occurrences With Constraints and Time Lags (MOW-CATL), facilitate knowledge discovery for sequential data mining problems that have groupings of events that occur close together, even if they occur relatively infrequently over the entire dataset. They work well for problems that have periodic occurrences when the signature of one sequence is present in other sequences, even when the multiple sequences are not globally correlated or spatially co-located. They also are able to handle a delay in time between the occurrence of the signature and the effect in the other sequences. Because of their flexibility, these data mining algorithms are well suited to handle knowledge discovery needs within the GDSS framework.; For the REAR approach, formal concept analysis is employed to develop the notion of frequent closed episodes from temporal data. The concept of representative association rules [35] is formalized in the context of event sequences. Constraints are used to target highly significant rules between infrequently occurring events. The REAR approach results in a significant reduction in the number of rules generated as compared to previous methods, while maintaining the minimum set of relevant association rules and retaining the ability to generate the entire set of association rules with respect to the given constraints.; MOWCATL is an efficient method for mining frequent sequential association rules from multiple data sets with a time lag between the occurrence of an antecedent sequence and the corresponding consequent sequence. This approach finds patterns in one or more sequences that precede the occurrence of patterns in other sequences, with respect to user-specified antecedent and consequent constraints.; The aim of this dissertation is to enhance the body of work in the area of data mining by using representative association rules [35], closures [56], constraints [61], and time lags in the context of event sequences. It also shows how these methods can enable knowledge discovery in the context of a GDSS, and provide examples of their application to the drought risk management problem.; *This research was supported in part by NSF Digital Government Grant No. EIA-0091530 and NSF EPSCOR, Grant No. EPS-0091900.