Research Of Bayesian Networks Classifier With Continuous Attributes

Bayesian Network is a graphical network data model according to probabilis-tic inference whose basement is Bayesian formula.Its main function is to gain probabilistic information of other variables by extracting information from cer-tain variables.Thus,Bayesian network is able to solve the problems about un-certainty and incompleteness.With this feature,Bayesian network classifier uses the marginalization decomposition and condition decomposition of probabilistic to compose Bayesian network in order to classify and predict data.Naive Bayesian Classifier is one of the most basic Bayesian networks whose precondition assumption makes it have the simple structure,highly efficient calcu-lation and fine classified effect,etc.However,the independence assumption of this condition have a marked impact on the accuracy of classify.Naive Bayesian Clas-sifier cannot consider the dependency relation between attribute variables which exactly plays the important role in classify.Based on it,this thesis makes the ex-tending research on Naive Bayesian Classifier by regarding continuous attribute as object and finally aims at improving the accuracy of classifier.The content of thesis is as follow:(1)It introduces the principle of Naive Bayesian Classifier with continuous at-tributes and estimates the related parameters by parameterization method and non-parameterization method.Then,it gives the details about several typical de-pendent expansion of naive Bayesian classifiers,and present different definitions for weight of Hidden Naive Bayesian Classifier.(2)It introduces Dynamic Naive Bayesian Classifier obtained by the combi-nation between Naive Bayesian Classifier and time series.Based on that,thesis proposes dynamic recessive naive Bayesian classifier by making the dependent ex-pansion and inducting a hidden parent node into attribute variable in every slice of time,and make corresponding inprovements.(3)It conduct numerical classification experiments for different classifiers and obtain the difference of classification effect between different classifiers by analyzing the experimental results.