Bayesian Network Structure Learning Based On MFO And Its Application

With the vigorous development of the information age,people accumulated mas-sive data,it is meaningful to learn the knowledge and logic behind the data and extract the valuable information that can be done by using Bayesian network.Bayesian net-work(BN)is a graphical network model which combines graph theory and probability theory,this mathematical tool is simple and clear,owing to its strong processing ca-pability for incomplete data and uncertain information,it is widely used in the data analysis,uncertainty reasoning and other fields,and worth promoting.The construc-tion of Bayesian network involves structure learning and parameters learning,in which structure learning is the key technology.The result of structure learning directly affects parameters learning,and then affects the application effect;therefore,the research on structure learning algorithms has a strong necessity.The main works of this dissertation are as follows:1.Considering the poor convergence,low accuracy of scoring-searching based structure learning algorithm which is the mainstream method and its easiness to fall into the local optima,BN-MFO(Bayesian Network Structure Learning using MFO)was proposed in this thesis and it is based on Moth-Flame Optimization.BN-MFO kept the sort framework of MFO,defined the crossover operator and variation operator by borrowing the ideas from genetic algorithm to replace the location update strategy of MFO.The mutual information between nodes was considered during mutation pe-riod to increase the possibility of returning a similar solution.Scoring-searching based structure learning algorithms contain a scoring metric and a search procedure,we con-ducted experiments to study the relations between two components of BN-MFO and the influence of R function on the convergence of BN-MFO,the validity of BN-MFO is analyzed.The results of the comparative experiments on the classical Cancer net-work and the Asia network shows that BN-MFO is generally superior to the same type of contrast algorithm,obviously,BN-MFO has strong superiority.2.Bayesian network was applied to the analysis of bank marketing data;this appli-cation involved all the major sections of Bayesian network research.BN-MFO was used for structure learning,the validity of this algorithm was tested in practical application.Since Bayesian network can be used for classification,we conducted an experiment with KNN and SVM to compare the classification accuracy,and the result of BN we built is better,which also reflects the correctness of BN-MFO.In order to facilitate the use of the model,the Matlab based GUI software was designed.In summary,in this dissertation we solved the common problems of scoring-searching based structure learning algorithms,and proposed a method to return the optimal struc-ture.The research of this thesis broadened the construction method of Bayesian network model,which has certain positive significance to promote the development of Bayesian network theory and broaden the application field of Bayesian network.