Abstract
The confirmation rule set concept, presented in this paper, provides a framework for reliable decision making. This framework enables flexible formation of a confirmation rule set, by incorporating rules induced by machine learning algorithms as well as human encoded expert rules. The only conditions for including a rule into the confirmation rule set are its high predictive accuracy and relative independence from other rules in the rule set. This paper introduces the concept of confirmation rule sets, together with an algorithm for selecting relatively independent rules from a set of acceptable confirmation rules. It presents also two approaches to confirmation rule induction: first, an exhaustive confirmation rule construction algorithm that was used to discover diagnostic rules in the coronary heart disease diagnosis problem, and second, a heuristic confirmation rule construction algorithm that was used for subgroup discovery in the coronary heart disease risk group detection problem.