Semi-supervised Bayesian network classifier learning based on inter-relation mining among attributes

Semi-supervised Learning as an efficient paradigm has been applied to many research areas, it also becomes one of the research focuses in machine learning and knowledge discovery. Traditionally, most classification models are built by supervised learning procedure, which leads to high rate of misclassification when test samples are significantly more than the training samples. This paper proposed to learn Bayesian classifier by using a semi-supervised procedure, which exploits the inter-relations among attributes mined from all test and training samples together to relax the conditional independent assumption of Naive Bayes(NB). Experimental results are presented to show the effectiveness and efficiency of the proposed approach.