Database dependency discovery: a machine learning approach

Article type: Research Article

Authors: Flach, Peter A. | Savnik, Iztok

Affiliations: Department of Computer Science, University of Bristol, Bristol BS8 1UB, UK E‐mail: [email protected] | Faculty of Computer and Information Science, University of Ljubljana, 1000 Ljubljana, Slovenia E‐mail: [email protected]‐lj.si

Abstract: Database dependencies, such as functional and multivalued dependencies, express the presence of structure in database relations, that can be utilised in the database design process. The discovery of database dependencies can be viewed as an induction problem, in which general rules (dependencies) are obtained from specific facts (the relation). This viewpoint has the advantage of abstracting away as much as possible from the particulars of the dependencies. The algorithms in this paper are designed such that they can easily be generalised to other kinds of dependencies. Like in current approaches to computational induction such as inductive logic programming, we distinguish between top‐down algorithms and bottom‐up algorithms. In a top‐down approach, hypotheses are generated in a systematic way and then tested against the given relation. In a bottom‐up approach, the relation is inspected in order to see what dependencies it may satisfy or violate. We give a simple (but inefficient) top‐down algorithm, a bi‐directional algorithm, and a bottom‐up algorithm. In the case of functional dependencies, these algorithms have been implemented in the FDEP system and evaluated experimentally. The bottom‐up algorithm is the most efficient of the three, and also outperforms other algorithms from the literature.

Keywords: Induction, attribute dependency, database reverse engineering, data mining

Journal: AI Communications, vol. 12, no. 3, pp. 139-160, 1999