Mining relational databases

Abstract

In this paper, we propose a classification system to induce an intentional definition of a relation from examples, when background kno- wledge is stored in a relational database composed of several tables and views. Refinement operators have been defined to integrate in a uniform way different induction tools learning numeric and symbolic constraints. The particularity of our approach is to use integrity constraints over the database (keys and foreign keys) to explore the hypotheses space. Moreover new attributes can be introduced, relying on the aggregation operator "group by".