A NEW ALGORITHM FOR INDUCTIVE LEARNING

This paper describes RULES-3, a new algorithm from the 'RULES' family of automatic rule extraction systems. These simple inductive learning systems for producing general rules from a collection of examples have a number of advantages over well known induction schemes. The immediate predecessor of RULES-3 had the following features: it could handle large sets of examples without having to break them up into smaller subsets, produced rules containing only relevant conditions, allowed a degree of control over the number of rules extracted, and could be applied to problems involving incomplete examples and objects with numerical attributes. RULES-S has two new features: it generates a compact set of more general rules and provides the user with the option of adjusting the precision of the extracted rules. The paper gives an example to illustrate the operation of the algorithm step by step and presents the results of evaluating it against its predecessor on the IRIS data classification problem.