Evaluation Of Bayesian Networks Based On Inter.iamb-Tabu Hybrid Algorithm And Its Application In Analyzing The Relating Factors Of Hyperlipidemia

Objective:In order to improve the performance of Bayesian network based on MMHC hybrid algorithm,three hybrid algorithms,MMPC-Tabu,Fast.iamb-Tabu and Inter.iamb-Tabu,are proposed to explore the effects of these three algorithms on building Bayesian networks.Compared with the MMHC hybrid algorithm;using the survey data of hyperlipidemia and its risk factors in Shanxi Province in 2013,the detection rate of hyperlipidemia in Shanxi Province was studied,and the optimal hybrid algorithm was used to construct the hyperlipidemia shell.The Bayesian network model studies the factors and correlation strengths directly or indirectly related to hyperlipidemia,and studies the possibility of occurrence of hyperlipidemia through model reasoning,providing a more reasonable method for the study of chronic diseases and related factors.Methods:Firstly,two kinds of standard Bayesian networks are used to randomly generate simulation data of different sample sizes,and then the Bayesian network is constructed by MMHC,MMPC-Tabu,Fast.iamb-Tabu and Inter.iamb-Tabu hybrid algorithms respectively.The effects of the Bayesian network constructed by the four algorithms are evaluated by the inverse edge,the missing edge,the number of redundant edges and the sum of them;the multi-stage stratified random sampling is used for 8 investigation points in Shanxi Province.Based on the survey data of hyperlipidemia and related factors,the multivariate logistic regression was used to screen the variables.The Bayesian network was constructed by the optimal hybrid algorithm,and the maximum likelihood estimation was used to estimate the parameters of the nodes in the network.Two models were compared to reflect the effects of variables on disease.Results:(1)Using two standard Bayesian networks to randomly generate data sets of different sample sizes,Bayesian network results constructed by four algorithms show that no matter how many the nodes and the sample size are,The more the Bayesian network structure is consistent with the standard network;the Bayesian network constructed by the Inter.iambTabu algorithm has higher consistency with the original network regardless of the number of nodes and the sample size.(2)The detection rate of hyperlipidemia in Shanxi Province in 2013 was 42.6%,and the 95% confidence interval was 41.1%~44.1%.The logistic regression model showed that gender,age,urban and rural,fresh vegetable intake,physical activity,BMI,central obesity,hypertension,and diabetes were associated with hyperlipidemia.Among them,central obesity has the highest correlation with hyperlipidemia,the OR value is 1.688,followed by BMI,and the OR value is 1.473.(3)The Bayesian network model of hyperlipidemia was constructed using the Inter.iamb-Tabu algorithm.The results showed that gender,age,physical activity,BMI,diabetes and hypertension were directly related to hyperlipidemia.Among them,gender,BMI,physical activity and age are the parents of hyperlipidemia,that is,they are related to the occurrence of hyperlipidemia;and diabetes and hypertension are the children of hyperlipidemia,Which means that hyperlipidemia is related to the occurrence of diabetes and hypertension;regional,central obesity,and intake of fresh vegetables are indirectly related to hyperlipidemia.(4)Bayesian network can also infer the probability of an unknown node(hyperlipidemia)according to the state of a known node,and determine the risk of hyperlipidemia.The results showed that the risk of hyperlipidemia was 0.455 if a person lacked exercise,0.553 if he had diabetes,0.647 if he had BMI > 28.0kg/m2,and 0.649 if he was central obesity according to waist circumference.Conclusion:1.The performance of the Bayesian network constructed by the Inter.iamb-Tabu hybrid algorithm is relatively ideal.2.The detection rate of hyperlipidemia in Shanxi Province was 42.6%,which was higher than the national level.3.The Bayesian network can not only reveal the factors and correlation strengths directly related to hyperlipidemia,but also reveal the indirect factors related to hyperlipidemia.Therefore,it is more targeted in disease prevention;it is directly related to hyperlipidemia.Among the factors,Bayesian network can identify the factors related to the occurrence of hyperlipidemia,and can identify the relationship between hyperlipidemia and other diseases.The Bayesian network's reasoning process is sequential,so if there are some defects,the reasoning of Bayesian network is still feasible.