The Research And Application Of Hidden Na?ve Bayes Based On Adaptive Attribute Selection

Na?ve Bayes Classifier(NBC)has been widely used because of the simple structure and the high efficiency of calculation,but it can't make full use of the relationship of the attributes due to the unrealistic condition independent assumption that affects its application.Hidden Na?ve Bayes Classifier(HNBC)reduces the restriction of independence by creating a hidden parent node for each attribute which combines the influences from all other attributes.And HNBC has better influence in classification accuracy than NBC.However,HNB has imperfection that redundant attributes affect classification accuracy,due to lacking selection mechanism of attribute-set.Base on this,our research work focuses on the following aspects:1.In this paper,we summarize the three existing attribute selection methods based on mutual information and propose a novel attribute selection model:adaptive maximum correlation attribute selection(AMaxMI).It applies equal-width discretization technology on discretizing mutual information value of all attributes,and then automatically chooses effective attributes and eliminates redundant attributes for classification according to the result of discretization.2.In this paper,we summarize the three existing attribute selection methods based on mutual information and propose a novel attribute selection model:adaptive enhance class correlation attribute selection(AMaxMI/MinR).This method regards mutual information of class-attributes as a reference index,and makes the discretization result of the reference index as the basis of selecting effective attribute automatically.3.In this paper,we propose two improved Hidden Na?ve Bayes classification algorithms by combining two kinds of adaptive attribute selection methods with HNBC:adaptive maximum correlation attribute selection of Hidden Na?ve Bayesian classifier(AMaxMI-HNBC),adaptive enhance class correlation attribute selection of Hidden Na?ve Bayesian classifier(AMaxMI/MinR-HNBC).And the simulation experiments show that the two kinds of improved Hidden Na?ve Bayes classification algorithms have higher accuracy and lower time complexity than Hidden Na?ve Bayes classification algorithm.4.We test thermal data of rotary kiln after preprocessing and then use the two kinds of improved algorithms above,and the experimental results show that the performance of the two methods is better than Hidden Na?ve Bayes.It reduces number of classification attribute,improves classification accuracy of Hidden Na?ve Bayes and reduces the computing time at the same time.