The experiments carried out in this paper suggest that a simple and quite naive best rule strategy performs quite well in comparison with the two other more complex strategies tested. As for PROSPECTOR-like approach, the results on Lymphography were quite bad, and similar to the best rule on the other datasets. With respect to MYCIN's certainty factors the performance was quite disappointing altogether.
The results did not show a clear advantage of the two traditional methods which combine different opinions. A possible cause for this could be a small number of rules to combine. This could perhaps improve if flexible matching were used.
The differences in classification accuracy observed between three different combination strategies indicate that more care should be taken when discussing the performance of learning systems. A great deal of work done in the area of approximate reasoning and uncertainty management could be exploited by the ML community.
Acknowledgements
I would like to thank both Pavel Brazdil and Zdenek Kouba for their help and comments.
REFERENCES
1. Bergadano, F., Matwin,S., Michalski,R., Zhang,J. : "Measuring Quality of Concept Descriptions", in EWSL88 - European Working Session on Learning, Pitman, 1988.
2. Brazdil,P., Gams,M., Sian,S., Torgo,L., Van de Velde,W. : "Learning in Distributed Systems and Multi-Agent Environments", in Machine Learning - EWSL91 ,European Working Session on Learning, Kodratoff,Y. (Ed), Lecture Notes on Artificial Intelligence, Springer Verlag, 1991.
3. Brazdil,P., Torgo,L. : "Knowledge Acquisition via Knowledge Integration", in Current Trends in Knowledge Acquisition, IOS Press, 1990.
4. Shafer, G. : A Mathematical Theory of Evidence, Priceton University Press, Princeton, 1976.
5. Duda,R., Hart,P., Nilsson,N. : "Subjective Bayesian methods for rule-based inference systems", in Proceedings of the AFIPS National Computer Conference, vol. 47, pp. 1075-1082.
6. Gams,M., Bohanec,M., Cestnik,B. : "A Schema for Using Multiple Knowledge", Josef Stefan Institute, 1991.
7. Heckerman,D. :"Probabilistic interpretation for MYCIN's certainty factors", in Uncertainty in Artificial Intelligence, Kanal,L. et al.(eds.), North-Holland, 1986.
8. Kouba,Z. : "Data Analysis and Uncertainty Processing", in Advanced Topics in Artificial Intelligence, Marik,V. et al. (eds.), Lectures Notes in Artificial Intelligence, Springer-Verlag, 1992.
9. Mántaras,R., : Approximate Reasoning Methods, Ellis Howood Limited, 1990.
10. Pearl, J. : Probabilistic Reasoning in Intelligent Systems: Networks of Plausible Inference, Morgan Kaufmann Publishers, 1988.
11. Shortliffe,E., Buchanan,B., "A Model of Inexact Reasoning in Medicine", in Mathematical Biosciences, 23, 1975.
12. Torgo,L. : YAILS an incremental learning program, LIACC-ML Group, Internal report nº 92.1, 1992.
13. Torgo,L. : "Controlled Redundancy in Incremtnal Rule Learning", in this volume.