A new approach for modelling uncertainty in expert systems knowledge bases

Abstract

The current paradigm of modelling uncertainty in expert systems knowledge bases using Certainty Factors (CF) has been critically evaluated. A way to circumvent the awkwardness, non-intuitiveness and constraints encountered while using CF has been proposed. It is based on introducing Data Marks for askable conditions and Data Marks for conclusions of relational models, followed by choosing the best suited way to propagate those Data Marks into Data Marks of rule conclusions. This is done in a way orthogonal to the inference using Aristotelian Logic. Using Data Marks instead of Certainty Factors removes thus the intellectual discomfort caused by rejecting the notion of truth, falsehood and the Aristotelian law of excluded middle, as is done when using the CF methodology. There is also no need for changing the inference system software (expert system shell): the Data Marks approach can be implemented by simply modifying the knowledge base that should accommodate them. The methodology of using Data Marks to model uncertainty in knowledge bases has been illustrated by an example of SWOT analysis of a small electronic company. A short summary of SWOT analysis has been presented. The basic data used for SWOT analysis of the company are discussed. The rmes_EE SWOT knowledge base consisting of a rule base and model base have been presented and explained. The results of forward chaining for this knowledge base have been presented and critically evaluated.