Intelligent computation for association rule mining

Although there have been several encouraging attempts at developing SQL-based methods for data mining, simplicity and efficiency still remain significant impediments for further development. In this paper, we develop a fixpoint operator for computing frequent itemsets and demonstrate three query paradigm solutions for association rule mining that use the idea of least fixpoint computation. We consider the generate-and-test and the frequent-pattern growth approaches and propose an novel method to represent a frequent-pattern tree in an object-relational table and exploit a new join operator developed in the paper. The results of our research provide theoretical foundation for intelligent computation of association rules and could be useful for data mining query language design in the development of next generation of database management systems.