Research On Bayesian Classifier Under The Supervised Learning

Bayesian classifier is a classifier algorithm set up on the bayesian statisticstheorem and bayesian network theorem. The bayesian classifier has the solidmathematics foundation and can be explained by the model easily. Therefore, it hasbeen the hot topic among the classification algorithms. In this paper, we studied thesupervised bayesian classifier. Through giving the advantages and disadvantages ofna?ve bayeian classifier, three new supervised bayesian classification models wereproposed.First, the existent bayesian incremental model has two main disadvantages: it existthe over fitting and not considering the added example's influence on the training setproblems. Therefore, an incremental algorithm based on the score function wasproposed in this paper. In this method, the score function can effectively solve the overfitting problem, and the proposed network equation considered how to choose theoptimal incremental examples. Moreover, this algorithm was used in the tree augmentedna?ve bayesian classifier, extending the usage.Next, choosing the attributes reasonably can improve the classificationperformance in the bayesian classifier. In this paper, we use the idea of attribute joiningin the BSEJ and unite the attributes which have the independence relationship by themutual information theory. Meanwhile, the different attributes have different influenceon the decision attribute. This means that different attributes should have differentweights. Therefore, the weighted na?ve bayesian classifier algorithm based on theconstructive induction was proposed.Last, the bayesian classifier model was combined with the integrated technology inthis paper. First, the K bayesian classifier model was improved as the unstable algorithm,then put forward the K bayesian classifier algorithm based on the Adaboost.