Learning Bayesian Network Classifier Based on Dependency Analysis and Hypothesis Testing

Abstract

Three types of representative Bayesian classifiers are analyzed and a new way to learn Bayesian Network classifier is introduced in this paper. At present, mutual information is frequently used to estimate whether variables independent of each other or not when constructing the Bayesian Network classifier structure, while this method has shortcomings of lacking of theoretical basis, which leads to low reliability. To improve this situation, a novel method Based on dependency analysis combined with hypothesis testing to form the structure of a Bayesian Network classifier is proposed here, and the junction tree algorithm of exact inference is adopted to form Bayesian Network classifier. The experimental results show that our approach is able to learn Bayesian Network classifier effectively, and enables Bayesian Network classifier to show a better performance in terms of classification accuracy than Naive Bayesian classifier and TAN classifier.