A KDD framework to support database audit

Information Technology and Management 1 (2000) 195–207 195A KDD framework to support database auditJean-Franc ¸ois BoulicautInstitut National des Sciences Appliquees´ de Lyon, LISI Batˆ iment 501,F-69621 Villeurbanne Cedex, FranceE-mail: jean-francois.boulicaut@insa-lyon.frUnderstanding data semantics from real-life databases is considered following an auditperspective: it must help experts to analyse what properties actually hold in the data andsupport the comparison with desired properties. This is a typical problem of knowledgediscovery in databases (KDD) and it is specified within the framework of Mannila andToivonen where data mining consists in querying theories e.g., the theories of approximateinclusion dependencies. This formalization enables us to identify an important subtask tosupport database audit as well as a generic algorithm. Next, we consider the DREAMrelational database reverse engineering method and DREAM heuristics are revisited withinthis new setting.Keywords: data mining, integrity constraint, reverse engineering1. IntroductionWe are interested in understanding data semantics from real-life databases. Thisprocess is considered following an audit perspective in the following sense: it must helpexperts to analyse what properties actually hold in the data and support the comparisonwith desired properties. This research paper takes examples from relational databaseaudit, assuming that inclusion and functional dependencies that ...