SubTree Augmented Naïve Bayesian classifier based on the fuzzy equivalence partition of attribute variables

Abstract

To make the structure of attribute variables in Naïve Bayesian classifier (NB) or Tree Augmented Naïve Bayesian classifier (TAN) more flexible and improve the accuracy of classification, a new Bayesian classifier called SubTree Augmented Naïve Bayesian classifier (STAN) is proposed in this paper. It adopts the fuzzy equivalence partition approach to partition attribute variables into several subsets and admits the structure of attribute variables to be several subtrees. NB and TAN can be easily simulated by STAN as the threshold changes. Experiments with UCI datasets and synthetic datasets demonstrate STAN is effective and efficient.