Theoretical Research Of Bayesian Network Learning Based On Mutual Information Variable Sequence Model

Bayesian network now has become a hot research field of artificial intelligence,is a product of the combination of probability theory and graph theory.In the research of Bayesian network structure learning,is an important branch of learning Bayesian network structure learning,and in the method,K2 algorithm is an important method of learning Bayesian network structure.K2 algorithm is a typical algorithm of search and evaluation,is currently one of the most widely used algorithm.In recent years,with the continuous development of intelligent optimization based on this algorithm,several extended algorithms have been produced.But because the dependent variables of the K2 algorithm,resulting in convergence and optimization algorithm has a great chance.So this study proposes whether the initial variables selection will be mutual information as the selection condition variables,so that you can select a variable sequence score higher to improve the efficiency of the construction of Bayesian networks,namely K2 sequence selection algorithm based on mutual information.Assume an arbitrary variable sequence can be seen as a linear type of BN network.The causal relationship between the linear network between each variable and correlated more strongly,more can reflect the relationship between the sequence of nodes.The causal relationship between the variables through the BN network evaluation function to measure,can be measured by the correlation between the mutual information between nodes.A new evaluation function is therefore to construct a evaluation based on BN network function and mutual information,as the evaluation measure of the merits of the variable node sequence function.For the above hypothesis,this study proposes a new scoring function variables,and joined the A,the B two weight coefficients,based on the original algorithm,to adjust the variables between the scores and the proportion of mutual information,to find the value of the variable structure high score better sequence.Based on mutual information as a new constraint,the proposed method,to solve the accuracy of Bias network construction under the condition of big data,data classification and application of Bias network inference and prediction data mining process to provide theoretical support and guarantee.The results show that the magnitude of the ordering score has a positive correlation with the score of the Bayesian network.With the increase of the ordering score the network scoring value of the Bayesian network is also increasing,so the new improved algorithm has an optimized effect on the structure of the new Bayesian network.Moreover,the sequence selected by the K2 sequence selection algorithm based on mutual information has a better structure than the original K2 algorithm,and improves the efficiency of building the Bayesian network,to a certain extent,especially in the case of large data sets.