Research On Bayesian Network Construction Algorithm Based On Combined Data

In recent years,Bayesian network has become one of the most popular models for expression and reasoning of uncertain knowledge.In the traditional Bayesian network learning algorithm,in order to obtain the scoring function of the dependency relationship between the nodes,the algorithm relies on the exponential level traversal process of the nodes,which causes the high time complexity and limits the training and learning of big data sets,especially in the case of complex data sources and the combination of data from different domains to build a Bayesian network.Considering complex data source and the combination of data from different fields to construct a Bayesian network will lead to a situation with many nodes,the paper proposes an algorithm of Bayesian network construction based on combined data.Firstly,we learn Bayesian network of different fields,and then fuses with a fusion algorithm to solve the problem of low efficiency caused by too many nodes.In addition,the paper use K2 to learn local Bayesian network.However,since scoring function is used to score every structure,K2 algorithm has high computational cost and time complexity.To solve this problem,an improved K2 is proposed in this paper.The new algorithm reduce calculation and improve algorithm efficiency by reducing using times of scoring function.This paper selects macro factors,stock and real estate related data to experiment on the proposed algorithm.Experimental results reveal:(1)the fusion algorithm is feasible;(2)the improved algorithm can reduce the computational resource and time complexity;(3)the efficiency and the ability that process large dataset of the fusion algorithm can be improved,compared with traditional K2 algorithm.