Rules + Exceptions: Automated Discovery of Comprehensible Decision Rules

Automated discovery of decision rules is a research area of significant importance as the discovered rules improve the decision making process in various real world situations across a wide spectrum of application fields. Rough set framework proposes automated discovery of decision rules and are particularly good at handling vagueness and uncertainty inherent to decision making situations. Though rough set theory discovers the high level symbolic decision rules (If-Then Rules) which are comprehensible individually, it produces large number of decision rules even for small datasets. A large set of rules may give high predictive accuracy but it is not comprehensible in the sense that it fails on the important criteria of manual inspection to gain insight into the application domain. This paper proposes a post processing scheme that takes the rules produced by rough set theory, organizes and summarizes the rules in the form of rule + exceptions structure consisting of default/general rules and their corresponding exceptions. The proposed scheme not only suitably organizes the decision rules for manual inspection and analysis, it also makes them more accurate and interesting by discovering exceptions.