| Bayesian network is an effective and versatile model learning tool that combines graph theory and probability theory to achieve a deeper and intuitive representation of joint probability distributions,thus learning potential models from data and further helping people to accomplish reasoning and decision-making tasks for uncertain problems in various fields.This paper has important practical applications in building Bayesian network structure learning algorithms,feature filtering and Bayesian gene networks.In this regard,this paper will carry out the following three aspects of discussion.(1)A new Bayesian network structure learning algorithm that incorporates integrated learning and frequent item mining is proposed.First,the original data set is sampled by Bootstrap to obtain the sample set,and the Apriori algorithm is used to mine the maximum frequent terms and association rules to determine the black and white list.Then the penalty term of the scoring function is constructed based on the black and white list,and the initial network is obtained by the hill-climbing search algorithm.The above steps are repeated to obtain multiple initial networks.Finally,these initial networks are fused into the final Bayesian network with the help of the integrated policy function.The Bayesian network structure learning algorithm fused with integrated learning and frequent term mining is compared with the BDEu algorithm on six large,medium and small standard networks,and it can be seen that the EF-BNSL algorithm can effectively improve the !! score,reduce the Hamming distance,and learn a network structure closer to the real network(2)The use of machine learning for diagnosis of NAFLD is a widely popular method.Part 2 of this paper applies Bayesian networks for feature screening and proposes a machine learning Voting algorithm for the detection and diagnosis of NAFLD.For a dataset containing 10508 medical examination results,the nearest neighbor algorithm is used to fill in the missing values.A Bayesian network was used for feature screening and 11 feature variables were retained.We selected four machine learning algorithms,including genetic algorithm,neural network,random forest and logistic regression,from 10 common machine learning algorithms with the highest diagnostic accuracy.These 4 methods were then used to become Voting algorithm using soft voting algorithm set.The results show that the proposed Voting algorithm outperforms other methods.(3)The construction of gene networks has significant applications for understanding the correlation between genes and finding pivotal genes,and Bayesian networks have shown good performance in constructing gene networks.In part 3 of this paper,we use colon cancer data as an example to construct gene networks using Bayesian networks.Differential genes were obtained by differential analysis.Inflammation-related genes were obtained from the Gene Cards database and intersected with differential genes to obtain 90 inflammation-related differential genes for functional enrichment analysis.The protein interaction network as well as Bayesian gene network were constructed,and it was found that the up-and down-regulated genes clustered significantly in the Bayesian gene network.The top 10 pivotal genes were selected and taken to intersect with survival-related genes,and finally 6 genes were obtained.Risk scores were calculated using multifactorial Cox and divided into high and low risk groups with median to do survival analysis,and the results showed that the survival probability was higher in the low-risk group. |
