GAB: Graph Augmented Bayes Classifier

This paper proposes a new classification approach; we call the graph augmented Bayes classifier (GAB). We show that naive Bayes classifier is a special case of GAB under the conditional independence assumption. GAB relaxes the conditional independence assumptions and takes into account of the influences on an attribute from all other attributes, and extends naive Bayes with the capability in expressiveness of non-linearly separable concepts. We conduct experiments by using datasets from the University of California at the Irvine repository. The experimental results show that the classifier extends naive Bayes with significant improvement in accuracy.