DISCOVERING BAYESIAN NETWORKS IN INCOMPLETE DATABASES
Dr Marco Ramoni, Knowledge Media Institute, The Open University
Dr Paola Sebastiani, Department of Actuarial Science and Statistics, City University

Abstract: A BBN is defined by a graphical structure of conditional dependencies among domain variables and a set of probability distributions defining these dependencies. During the past few years, several efforts have been addressed to develop methods able to extract both the graphical structure and the conditional probabilities of a BBN from a database. All these methods share the assumption that the database at hand is complete, that is, it does not report any entry as unknown. When this assumption fails, these methods have to resort to expensive iterative procedures which are infeasible for large databases. This talk will introduce an efficient method to estimate conditional probabilities from incomplete databases called Bound and Collapse (BC) and it will show how it can be used to extract the graphical structure of a BBN from an incomplete database. Experimental comparisons with other estimation methods, such as Gibbs Sampling and EM algorithm, will be reported and Bayesian Knowledge Discoverer (BKD), a computer program implementing BC, will be demonstrated using a large real-world medical database. Further information is available from the BKD site at http://kmi.open.ac.uk/~marco/projects/bkd.

This seminar was held at the Department of Computer Science, Royal Holloway, University of London on 19 November 1997.

back