摘要： 针对朴素贝叶斯模型的稳定性，进一步提高朴素贝叶斯模型的性能，通过集成学习方法克服朴素贝叶斯模型中属性独立的限制条件，提出一种基于Oracle选择的朴素贝叶斯集成算法，使用Oracle选择机制破坏其稳定性，并从中选取较好的分类器作为集成学习中的个体成员，使用投票方法对结果进行融合。实验结果证明，该算法能提高朴素贝叶斯模型分类的正确率，表明OSBE的性能在一些数据集上优于Bagging与Adaboost集成学习的性能。

关键词: 朴素贝叶斯集成, Oracle机制, 稳定性, 投票法

Abstract: Aiming at the stability of native Bayesian, in order to improve the performance of the Native Bayesian, and to overcome the limitation of the attributes independence assumption in the native Bayesian with ensemble, this paper presents an algorithm for building ensembles of native Bayesian classifiers on oracle selected. In this algorithm. It makes the stability of the native Bayesian with Oracle weaken, then selects the better classifier to be as the number of ensemble of the native Bayesian classifiers, and integrates the classifiers with voting method. Experimental result proves the OSBE ensemble algorithm obviously improves the accuracy of the native Bayesian. And it proves in some cases the OSBE ensemble algorithm has better classification accuracy than Bagging and Adaboost.

Key words: naive Bayesian ensemble, Oracle strategy, stability, vote method

中图分类号: 

• TP18