Research On High Dimensional Bayesian Network And Its Application In Colleges' Input-output Performance

As a powerful tool to solve the problem of uncertainty knowledge reasoning,Bayesian network has been widely used in data mining and artificial intelligence field.However,the traditional Bayesian network is difficult to be used for discovering the causal relationship with high-dimensional data for the following reasons.At first,the structure learning of bayesian network is a np-hard problem.the traditional bayesian network structure learning algorithm can always learn network structure well with a small number of nodes.However,as the nodes in the network increase,the network structure search space increase exponentially.Secondly,there is no causal semantics between network nodes,which can not reflect the causal relationship between nodes.Finally,when constructing Bayesian network on the high-dimensional data,the traditional mountain climbing method will affect the network reasoning performance because of the loss of potential edge in the network reduction and combination.To deal with these three problems,The main works of this paper are as follows:Firstly,add the causal relationship between the nodes based on the traditional mdl scoring function of mountain climbing method.then,This paper proposes a climbing algorithm based on causality score with the modified traditional edge reduction and turn operator.And the edges of the network is causally recognized before taking edge reduction and turn operator,so that,we can not only avoid accidental deletion of the causal edges,but also modifying the edge direction of parent node and child node in the original network by turn operator.By doing that,the network can reflect the causal relationship of data better and improve the network reasoning performance.Secondly,in order to reduce the influence of the loss of the potential side to the network reasoning performance,this paper divide the establishment of high-dimensional Bayesian network into multiple Bayesian subnets with the idea of divide and conquer.In addition,this paper define the similarity based on the causal measure for the purpose split the causal nodes in the same subnet.Then,the Bayesian subnet is established based on the causal score.Finally,compared with the traditional mountain climbing algorithm,the causal mountain climbing algorithm has advantages in network causality measurement ability,network score and reasoning performance.Finally,the causal mountain climbing algorithm is applied to the weak analysis of input and output in colleges and universities.Further more,for the purpose of exploring the influence when applying the reduced combination method,this paper compared to the bayesian networks constructed by the traditional mountain climbing algorithm and the causal mountain climbing algorithm.It is found that the causal mountain climbing algorithm not only in the network score is better than the traditional mountain climbing algorithm,but also in the complexity of the network remains unchanged at the same time,the network overall reasoning performance improved 11.7%.