基于贝叶斯网络的分类算法研究及其在CRM中的应用-计算机科学与技术专业论文.docxVIP

AbstiaetIn Abstiaet In the view of data mining,classifier problem is to establish a model for classification Or find rules，and assign the unclassified instance to any class．Bayesian classifier is researchers’favor for its solid theoretical foundation and its good 1,erformanee．Improved classifier based On Navie Bayesian classifier and learning Bayesian network structure is two aspects in the area of Bayes research．Tllis paper is focus 013 the former． Bayesian classifier is to find any class、航th maximum posterior probablity in the hypothesis space and assign the unclassified sample to it．Navie Bayesian dassifier(NB)is a classic bayesian classifier、7Iritll the hypothesis that all attributes a∞ condition independent．It is proved that NB has the ssmt。classify accuracy winl decision tree classifier and neural network classifier．In practice,there锄蟹certain relevance among somlo attributes．Its strict condition constraint im’t able to reflect the relationships and also lower its classification accuracy．Some improved algorithm is based Oil the relaxation of its constraint．Tree augmented navie bayesian classifier(TAN)is觚improved algorithm．It allows class node to be its father node and also allows a non-class node to be its father node at most．TAN classifier performs better than NB classifier in classification accuracy for classification in sonic samples． 111e role of every attribute is different for classification．So how to measure the different influence?Attribute橱etivity can be used to mea．观lle the relevance among attributes．Kinds of correlation analysis methods a托introduced in the paper,such as method based 011 statistics，method based On mutual information and SO Oil．Attribute reetivity analysis is used in many applications，for example attribute reduction based 011 attribute relevance,weighted model and attribute grouping．Except for the mentioned content，the weighted model based On these attribute analysis methods is also introduced． Another application for