Naïve Bayes ensemble learning based on oracle selection
Kai Li, Lifeng Hao · 2009
Aiming at the stability of Naive Bayes algorithm and overcoming the limitation of the attributes independence assumption in the Naive Bayes learning, we present an ensemble learning algorithm for naive Bayesian classifiers based on oracle selection (OSBE). Firstly we weaken the stability of the naive Bayes with oracle strategy, then select the better classifier as the component of ensemble of the naive Bayesian classifiers, finally integrate the classifiers' results with voting method. The experiments show that OSBE ensemble algorithm obviously improves the generalization performance which is compared with the Naive Bayes learning. And it prove in some cases the OSBE algorithm have better classification accuracy than Bagging and Adaboost.