A note on the application of classical statistics to evaluating the knowledge base of an expert system

The paper explores the use of classical statistical procedures for evaluating the knowledge base of an expert system that outputs quantitative belief values. Statistical procedures can be used to estimate the relative frequency that an expert system will output belief values which: a) strongly support a correct hypothesis, and b) do not provide strong support for any hypothesis. These procedures can also be used to estimate the minimum number of test problems required to evaluate an expert system. >