Daniel J. Stein III, Sheila B. Banks, Eugene Santos, Jr., and Michael L. Talbert
In this paper we present a methodology for goal-directed data mining of association rules and incorporation of these rules into a probabilistic knowledge base. The purpose of the data mining and rule extraction process is to repair knowledge base incompleteness uncovered during validation. We discuss how this incompleteness is uncovered and show the fundamental forms this incompleteness can take. We describe how association rules can be extracted from databases in order to address excluded information and to express missing relationships in a probabilistic knowledge base. The current implementation of this goal-directed data mining within an integrated generic expert system tool is also described. Our methodology can benefit many data intensive and imprecise domains such as stock market analysis, intelligence analysis, and operational management.