An efficient strategy for mining exceptions in multi-databases

This paper proposes a new strategy, referred to as local instance analysis, for multidatabase mining. While many interstate organizations have an imperative need to analyze their data in multi-databases distributed throughout their branches, traditional multi-database mining utilizes the strategies for mono-database mining: pooling all the data from relevant databases into a single dataset for discovery. This leads to the destruction of useful information, for instance, 70% of branches within a company agreed that a married customer usually has at least 2 cars if his/her age is between 45 and 65’. This information assists in global decision-making within the company. Our new strategy is developed for discovering this useful information. Using the local instance analysis, we design an algorithm for identifying exceptions from multi-databases. Exceptional pattern reflects the individuality’ of, say, branches of an interstate company. 2003 Published by Elsevier Inc. * Corresponding author. Address: Faculty of Information Technology, University of Technology, Sydney, P.O. Box 123,Broadway, NSW 2007, Australia. Fax: +61-2-9514-1807. E-mail addresses: [email protected] (S. Zhang), [email protected] (C. Zhang), [email protected] (J.X. Yu). 0020-0255/$ see front matter 2003 Published by Elsevier Inc. doi:10.1016/j.ins.2003.10.008 2 S. Zhang et al. / Information Sciences xxx (2003) xxx–xxx ARTICLE IN PRESS