Research And Application Of Bayesian Network Structure Learning Algorithm

Bayesian network is a probabilistic graphical model that can are capable of effective inference diagnosis and data analysis of uncertainty problems in complex domains.Parameter learning in Bayesian network learning is predicated on structure learning.As the number of network nodes increases,the likelihood of the network structure space grows exponentially.Learning Bayesian structures is NP-hard even when the maximum number of parents of each node in the network is limited to two,so it is of great theoretical and practical importance to study and find the optimal Bayesian network structure learning.The following studies have been conducted:Firstly,the research background,development status and advantages and disadvantages of existing Bayesian network structure learning methods are investigated.The improvement of hybrid bayesian structure learning algorithm can improve the accuracy and efficiency by constraining the search space through clustering in the initial stage is concluded through the collation and study of literature.Secondly,an idea approach based on clustering and integration is proposed to improve the Bayesian network structure learning algorithm.First,the clustering idea is used to generate a priori networks to improve the accuracy and efficiency of Bayesian network structure learning.Then the Bayesian network is further learned by combining the integration idea,and several sample sets are obtained by Monte Carlo sampling,and the Bayesian network is generated by combining the prior Bayesian network on different samples respectively,represented by the adjacency matrix to get the edge weights,and the integrated Bayesian network is obtained by threshold screening.Finally,the integrated network obtained as the prior network is updated and iteratively learned until the Bayesian network with the largest score is generated.In this paper,the?5₁ value,Hamming distance and the number of correct edges are obtained by comparing with different algorithms on a standard base dataset.The results show that our proposed clustering and integration-based Bayesian network structure learning algorithm yields a network with higher correctness and better performance.Thirdly,based on the proposed structure learning algorithm based on clustering and integration of Bayesian networks is applied on medical field.The effectiveness of the proposed algorithm on real data is verified by comparing it with other hybrid structure learning algorithms and other Bayesian classifier.19 pictures,8 tables,90 references.