Representation of propositional expert systems as partial functions

摘要

Propositional expert systems classify cases, and can be built in several different forms, including production rules, decision tables and decision trees. These forms are inter-translatable, but the translations are much larger than the originals, often unmanageably large. In this paper a method of controlling the size problem is demonstrated, based on induced partial functional dependencies, which makes the translations practical in a principled way. The set of dependencies can also be used to filter cases to be classified, eliminating spurious cases, and cases for which the classification is likely to be of doubtful validity.