Bayesian Network Classifiers And Application

Bayesian network is a graphics mode representing random variable dependenciesamong variables, a powerful tool for uncertainty knowledge representation andreasoning, combination of probabilistic and graphical theory, has been widely appliedin many fields. Based on Bayesian network, we can decompose condition and edge ofthe joint probability, reduce computational complexity consequently, and solve arange of issues related to joint probability calculation. Bayesian network forclassification prediction is often referred to as a Bayesian network classifier.Naive Bayesian classifier is a basis for a Bayesian network classifier, and isknown for simple, efficient and good classification accuracy. However, this classifieris based on a strong assumption of conditional independence, making the dependencyinformation between attributes cannot be used effectively, which is often an importantpart for classify. Starting from the expansion, optimization and application of NBC, Iresearch on selective Bayesian classifier(SNB), Tree Augmented Naive Bayesianclassifier(TAN), K-dependence augmented NBC, Bayesian network augmented NBC,Complete Bayesian classifier(CBC), Boosted NBC and Dynamic Naive Bayesianclassifier(DNBC).The main contents of this dissertation are as follows:(1) Introduced different ways to deal with continuous and discrete attributes,feature subset selection, and ways to estimation conditional probability density of theattributes based on Gaussian and Gaussian kernel function.(2) Described in detail the development of NBC, and analyzed several typicalaugmented NBCs(such as TAN, KDB, class-restricted Bayesian network classifiers)and pointed out the strengths and weaknesses of each classifier.(3) Given some applications examples of enterprise financial early warning,early warning of operational risk and university teachers’ capacity assessment basedon Naive Bayesian classifier and its augmented classifiers. (4) Studied Dynamic NBC by combining with NBC and time series, and given afeature subset selection method of DNBC, and verified the validity of this method. Atlast, I gave two applications on analyzing the impact of economic growth andcommodity import and export.