Active Database Rule Set Reduction by Knowledge Discovery

The advent of active databases enhances the functionality of conventional passive databases. A large number of applications benefit from the active database systems because of the provision of the powerful active rule language and rule processing algorithm. With the power of active rules, data manipulation operations can be executed automatically when certain events occur and certain conditions are satisfied. Active rules can also impose unique and consistent constraints on the database, independent of the applications, such that no application can violate. The additional database functionality offered by active rules, however, comes at a price. It is not a straightforward task for database designers to define and maintain a large set of active rules. Moreover, the termination property of an active rule set is difficult to detect because of the subtle interactions among active rules. This dissertation has proposed a novel approach of applying machine learning techniques to discover a set of newly simplified active rules. The termination property of the discovered active rule set is also guaranteed via the stratification technique. The approach of discovering active rules is proposed in the context of relational active databases. It is an attempt to assist database designers by providing the facility to analyze and refine active rules at designing time. The main algorithm of active rule discovery is called the ARD algorithm. The usefulness of the algorithm was verified by the actual running on sample sets of active rules. The running results, which were these corresponding new sets of active rules, will be analyzed on the basis of the size and the complexity of the discovered rule sets. The size of the discovered rule set was analyzed in term of the number of active rules. The complexity was analyzed in term of the number of transition states, which are the changes in the database states as the result of rule execution. The experimental results revealed that with the proposed approach, the numbers of active rules and transition states could be reduced 61.11 % and 40%, respectively, on average