贝叶斯分类器的判别式参数学习

Discriminative parameter learning of Bayesian network classifier

为了提高贝叶斯分类器的分类性能,针对贝叶斯网络分类器的构成特征,提出一种基于参数集成的贝叶斯分类器判别式参数学习算法PEBNC。该算法将贝叶斯分类器的参数学习视为回归问题,将加法回归模型应用于贝叶斯网络分类器的参数学习,实现贝叶斯分类器的判别式参数学习。实验结果表明,在大多数实验数据上,PEBNC能够明显提高贝叶斯分类器的分类准确率。此外,与一般的贝叶斯集成分类器相比,PEBNC不必存储成员分类器的参数,空间复杂度大大降低。

Concerning the characteristics of Bayesian networks classifier,a discriminative parameter learning algorithm of Bayesian networks classifier based on parameters ensemble named PEBNC was proposed to improve the classification performance of Bayesian classifier.This algorithm regarded the parameter learning as a regression problem,applied the additive regression model to the parameter learning of Bayesian networks classifier,and realized a discriminative parameter learning of Bayesian networks classifier.The experimental results indicate that the PEBNC classifier can improve the classification performance in most cases.Furthermore,compared with the general Bayesian classifier ensemble,PEBNC requires less space because there is no need to save parameters of individual classifiers.