Research Of Data Mining Based On Bayesian Network Structure Learning And Classifier

There are many uncertain problems in the real world and other scientific fields. Bayesian network is developed by the integration of probability with graph theory. It use graph theory to reveal the structure of the problem on one hand. On the other hand, it makes full use of the structure of the problem to reduce the complexity of the problem under probability theory. It provides a natural tool for dealing with problem of uncertainty, and Bayesian network has been used widely in many fields, for example, agriculture, medical treatment, finance, industry and government decision-making. Moreover it has acquired good social benefit and economy benefit, and it will have broad perspective of exploitive and application. So it is available to do research on it.The main contents of this thesis are as follows:First, the overview of Bayesian network. This thesis introduces and analyses the background and purpose of data mining and Bayesian network. It also summarizes the superiorities and characteristics that Bayesian networks compares with other methods. After discussing its function and reasoning mechanism, the thesis focus on the aims and major problems of Bayesian networks.Second, this thesis concretizes search and score algorithm and the idea of mutual information which are used for leaning structure of Bayesian network. It uses mutual information at the search period of the search and score algorithm and put forward an algorithm CIWK for leaning structure of Bayesian network. The algorithm uses mutual information to get the max weight spanning tree. Then, a method called chain structures is used to reduce the complexity. At last, we get an order of variables which used as an input parameter to the algorithm K2. The lower complexity is one of the superiorities and characteristics of this algorithm. Theoretical analysis and experimental results show algorithm CIWK has good performance.At last, the algorithm CIWK is used as a part of Bayesian network classifier. And the experiment indicates: classifier GBN is better than classifier BN and Rule, and is close to the classifier C4.5.