The Naïve Bayesian Classifier Learning Algorithm Based on Adaboost and Parameter Expectations

Naïve Bayesian classifier is a simple classification method based on Bayes statistics, which is one of the most popular classifiers and has been successfully applied to many fields. To improve the generalization ability of the naïve Bayesian classifier, discriminative learning of the naïve Bayesian classifier is researched. In this paper, a parameter learning algorithm AENB of the naïve Bayesian classifier is proposed. This algorithm adopts the Adaboost´s classifier ensemble framework, sequentially generates a series of individual classifiers with parameters, and obtains parameter expectations by summing the weighting parameters of each individual classifier. In the final, the naïve Bayesian classifier with parameter expectations is constructed. The experimental results show that the AENB algorithm improves classification accuracy of the naïve Bayesian classifier in the most cases. Furthermore, compared with the naïve Bayesian classifier ensemble, AENB requires less space because there is no need to save parameters of individual classifiers.