Research Of Bayesian Networks Classifier With Discrete Attributes

Bayesian network is a tool for modeling uncertainty.It represents the rela-tionship of random variables by graph model,which is a combination of probability theory and graph model.Bayesian network is a classification model.Classification learning is one of the major issues of machine learning and big data mining.At present,this theory and technique have been widely used in many fields.It can classify big data information efficiently and effectively according to the dependen-cy relationship among the attribute of variables.The Bayesian network used for prediction and classification is called Bayesian classification.According to the at-tribute of variables,the Bayesian classification can be divided into discrete Bayesian classification,continuous Bayesian classification,and mixed Bayesian classification.This paper studies discrete Bayesian classification.In order to improve the accuracy of classification,scholars and experts have proposed dozens of methods and theories of classification based on the dependency of the attributes nodes,such as Naive Bayes classification,Tree Augmented native Bayes,Hidden Naive Bayes,complete Bayesian classification and so on.The Hid-den Bayes classifier is the one of the most basic classification.It is premised on the assumption that the nodes attributed are independent of each other.This makes the dependency information between the attributes unable to be analyzed,and this part may contain some critical information.In order to improve the accuracy of the Hidden Bayesian classification and reduce the strict conditional independence of the attribute variables,Friedman proposed a new type of classification-the TAN classifier-to slack the assumption of independence of attribute nodes and allow at-tributes possess an attribute node as its attribute parent node,besides class node,so the tree augmented classification relies on the attribute node information partially.When the dependency between attribute nodes is strong,Hidden Bayes classifica-tion,K dependent attribute node classification,semi-hidden Bayes classification are proposed;besides changing the Bayesian model,we can also improve the accuracy of classification via altering the weights of attribute variables.The assumptions of attribute independence and equal importance are not consistent with objective facts sometimes,so the experts and scholars have proposed weighted Hidden Bayesian classification,which is based on feature items of mutual information.This paper proposes two improvements.Firstly,to improve the classification accuracy by altering the coefficient power of the weight in the Naive Bayes clas-sification.Secondly,to compare the connection direction of the edge of the tree augmented attribute classification in the implicit naive Bayes.According to the condition dependence theorem,to determine the connection direction of the edge between the attribute node and its corresponding hidden parent node.