Converting a Naive Bayes models with multi-valued domains into sets of rules

Abstract

Nowadays, several knowledge representation methods are being used in knowledge based systems, machine learning, and data mining. Among them are decision rules and Bayesian networks. Both methods have specific advantages and disadvantages. A conversion method would allow to exploit advantages of both techniques. In this paper an algorithm that converts Naive Bayes models with multi-valued attribute domains into sets of rules is proposed. Experimental results show that it is possible to generate rule-based classifiers, which have relatively high accuracy and are simpler than original models. © Springer-Verlag Berlin Heidelberg 2006.