Verification, Testing and Validation of Rule-Based Expert Systems

Abstract Expert systems (ES) are used for applications in automatic control requiring high reliability. A failure in such systems can cause a great disaster. We give an overview of validation, verification, and testing methods that can be used to improve ES reliability. The test data generation for rule-based ES is investigated in detail. The statement and contribution adequacy criteria for analysing and testing ES are introduced. These criteria may be applied to different types of rule-based expert systems. The problems offinding test sets for ES without uncertainties, for ES using one uncertainty measure, and for ES allowing fuzzy truth values are considered. Several classes of ES with polynomial static analysis and test data generation complexity are characterized.