ABSTRACT
Rule induction is the branch of machine learning that remedies to the mutually exclusivity of the rules generated by decision tree methods. The known rule induction methods (PVM, CHARADE) can be applied only to little data sets since they use costly procedures [WK91, DK91]. In this paper, we present a faster rule induction method, the Incremental Production Rule based method (IPR method). The empirical evaluation in many real world applications proves that IPR is more precise than the rule induction method PVM and the decision tree methods ID3 and SIPINA. Other empirical results focus on the usefulness of the rule conflict resolution algorithm used in the IPR system.
