摘要
NaveBayes分类器作为一种计算简单、精度较高的分类方法,已经得到了广泛应用。但是其所作的假设:各属性之间相互独立却非常容易在现实中被违背,阻碍了分类器精度的进一步提高。而Bayes网络较好地考虑了属性之间的依赖关系,但是其计算相当复杂。AugmentedBayes分类器将两者的优点结合在一起,既考虑了属性之间的依赖关系,又保证了算法的简单性。该文从属性所拥有的信息量出发考虑,提出了AugmentedBayes分类器的一种基于熵的学习方法。最后,通过测试数据将该方法与NaveBayes分类器和SuperParent算法进行了比较。
Nave Bayesian Classifier has been broadly in practice because of its efficient computation and good performance.But the assumption that all attributes are independent is easily violated in real world and it is an obstacle to refine the accuracy.Bayesian Network relaxes the assumption of independence,but its algorithm is very complex.Upon the consideration of the conditional dependence Augmented Bayesian classifier remains the efficient computation.This paper explores an entropy-based approach of learning augmented Bayesian classifier.Finally this method is compared to Nave Bayes and SuperParent through test data.
出处
《计算机工程与应用》
CSCD
北大核心
2002年第17期100-102,共3页
Computer Engineering and Applications