Quality Measures for Bayesian Network Classifier Design

Abstract: The paper is dedicated to classification of documents into one of available classes. The role of a classifier is played by Bayesian network classifiers having the structure of an augmented Naive Bayes classifier. The focus is on quality measures enabling to compare different Bayesian networks and select the better one while searching a space of possible network structures. Experiments with several quality measures and several types of network structures were carried out using an English document collection