Tailored-to-Fit Bayesian Network Modeling of Expert Diagnostic Knowledge
This paper addresses issues in constructing a Bayesian network domain model for diagnostic purposes from expert knowledge. Diagnostic systems rely on suitable models of the domain, which describe causal relationships between problem classes and observed symptoms. Typically these models are obtained by analyzing process data or by interviewing domain experts. The domain models are usually built in the forward direction, i.e. by using the expert provided probabilities of symptoms given individual causes and neglecting the information in the backward direction, i.e. the knowledge about probabilities of problems given individual symptoms. In this paper we introduce a novel approach for the structured generation of a model that incorporates as closely as possible that subset of the unstructured multifaceted and possibly conflicting probabilistic information provided by the experts that they feel most confident in estimating.
Keywordsexpert systems Bayesian networks for troubleshooting and diagnosis probabilistic model causal model noisy OR proxy model probabilistic information experts feel confident in estimating generation of a model that matches as closely as possible the probabilistic information provided by experts smallest forward-backward expert-based model model generation closest match
Unable to display preview. Download preview PDF.
- 1.D. Obradovic and R. Lupas Scheiterer, “Troubleshooting in GSM Mobile Telecommunication Networks based on Domain Model and Sensory Information,” ICANN, 2005.Google Scholar
- 2.J. Horn, T. Birkhölzer, O. Hogl, M. Pellegrino, R. Lupas Scheiterer, K.-U. Schmidt, and V. Tresp, “Knowledge Acquisition and Automated Generation of Bayesian Networks,” Proc. AIME ’01, Cascais, Portugal, 2001, pp. 35–39, July.Google Scholar
- 3.J. Horn, T. Birkhölzer, O. Hogl, M. Pellegrino, R. Lupas Scheiterer, K.-U. Schmidt, and V. Tresp, “Knowledge Acquisition and Automated Generation of Bayesian Networks for a Medical Dialogue and Advisory System,” in Artificial Intelligence in Medicine, S. Quaglini, P. Barahona, and S. Andreassen (Eds.), Springer, 2001, pp. 199–202.Google Scholar
- 4.F. V. Jensen, “An Introduction to Bayesian Networks,” UCL Press, 1996.Google Scholar
- 5.K. Murphy, “Software Packages for Graphical Models/Bayesian Networks,” last updated 31 October 2005 (status at paper submission), http://www.cs.ubc.ca/~murphyk/Bayes/bnsoft.html.
- firstname.lastname@example.org, “Bayesian Network Repository,” http://www.cs.huji.ac.il/labs/compbio/Repository/, March 01, 2001 (status at paper submission).
- 7.J. Pearl, “Probabilistic Reasoning in Intelligent Systems, Morgan–Kaufmann,” 1988.Google Scholar
- 8.D. Heckerman, “A Tutorial on Learning With Bayesian Networks, Technical Report,” http://research.microsoft.com/~heckerman, March 1995.
- 9.J. Kim and J. Pearl, “A Computational Model for Causal and Diagnostic Reasoning in Inference Engines,” Proceedings Eighth International Joint Conference on Artificial Intelligence, Karlsruhe, West Germany, pp. 190–193.Google Scholar
- 10.M. Henrion, “Some Practical Issues in Constructing Belief Networks,” Proceedings of the Third Workshop on Uncertainty in Artificial Intelligence, Seattle, WA, Association for Uncertainty in Artificial Intelligence, Mountain View, CA, pp. 132–139 (also in “Uncertainty in Artificial Intelligence,” L. Kanal, T. Levitt, and J. Lemmer (Eds.), North-Holland, New York, 1989, pp. 161–174.Google Scholar
- 11.D. Heckerman, “Causal Independence for Knowledge Acquisition and Inference,” Proceedings of the 9th Conference on Uncertainty in Artificial Intelligence, Washington, DC, Morgan Kaufmann, San Mateo, CA, 1993, pp. 122–127.Google Scholar
- 12.R. Lupas Scheiterer, “HealthMan Bayesian Network Description: Disease to Symptom Layers, Multi-Valued Syptoms, Multi-valued Diseases,” Siemens AG Internal Report, October 1999.Google Scholar
- 13.R. Lupas Scheiterer, “Bayesian Network Modeling Aspects Resulting from the HealthMan and GSM Troubleshooting Applications,” Siemens AG Internal Report, April 2003. Google Scholar