Learning Conditional Probabilities from Incomplete Databases - An Experimental Comparison

Authors:

Marco Ramoni
Knowledge Media Institute
The Open University
Milton Keynes, MK7 6AA
United Kingdom
E-mail: m.ramoni@open.ac.uk
Phone: +44 (1908) 655721
Fax: +44 (1908) 653821

Paola Sebastiani
Statistics Department
The Open University
Milton Keynes, MK7 6AA
United Kingdom
E-mail: p.sebastiani@open.ac.uk
Phone: +44 (1908) 652359
Fax: +44 (1908) 652140

Abstract:

This paper compares three methods - the EM algorithm, Gibbs sampling, and Bound and Collapse (BC) - to estimate conditional probabilities from incomplete databases in a controlled experiment. Results show a substantial equivalence of the estimates provided by the three methods and a dramatic gain in efficiency using BC.

Keywords:

Missing data, EM algorithm, Gibbs sampling, Bound and Collapse.

Availability:

Postscript and PDF.

Other information:

Futher information is available from the home page of the Bayesian Knowledge Discovery project at The Open University.