Application Of Uric Acid Related Factors Based On Continuous Bayesian Network

Objective:To study the continuous bayesian network structure learning algorithm based on partial correlation and compare with the discrete structure learning algorithm MMHC.Aiming at the problem that the discretization of continuous variables may lead to the loss of information,the simulation data is used to test the ability of IPCB continuous structure learning algorithm to make full use of the data information to construct the network framework accurately.Use chronic disease investigation data in 2015,Shanxi Province,MMHC and IPCB algorithm is adopted to establish the uric acid and its related factors of bayesian network structure respectively.The IPCB algorithm can make full use of the information provided by the continuous variable data to establish a more complete network of disease-related factors.This study can provide new ideas for chronic diseases influencing factors of network analysis.Methods:In GeNIe2.4,the test network was selected to generate continuous data sets and discrete data sets with different sample sizes.IPCB algorithm and MMHC algorithm were respectively used to learn the bayesian network structure.Compared with the original network structure,the error edge number was used as the index to compare the performance of the algorithm.Based on the monitoring data of chronic diseases and nutrition in Shanxi Province in 2015,a data set composed of continuous variables that may be related to uric acid was selected for analysis.According to the research content,the data set was divided into metabolic index data set and dietary adjustment index data set.First a statistical description of two data sets are simple,and then the IPCB algorithm is adopted to establish the uric acid in the two data set related factors of continuous variable bayesian networks,at the same time after the variable discretization MMHC algorithm is adopted to establish the corresponding discrete bayesian network,and compare with continuous bayesian network,compare two algorithms to establish the rationality of uric acid related factors of bayesian networks.Results:(1)through the test of ASIA and TANK simulation networks,it is found that the learning bayesian network based on MMHC algorithm based on variable discretization can obtain some correct edges,but the missing edges are prominent regardless of the sample size.However,IPCB continuous structure learning algorithm found in the test that IPCB algorithm could obtain the correct and complete network structure regardless of the sample size.(2)the continuous variable index related to uric acid was extracted from the database of chronic diseases and nutrition monitoring in Shanxi Province in 2015,and the database was divided into two data sets,namely the data set of metabolic index and the data set of catering households,according to the research content.In the data set of metabolic indicators,8 consecutive variables including uric acid,triglyceride,total cholesterol,LDL cholesterol,HDL cholesterol,fasting blood glucose,hba1 c and age were selected.Body measurements and dietary indicators related to uric acid: diastolic blood pressure,systolic blood pressure,BMI,waist circumference,meat intake,aquatic product intake,salt intake,edible oil intake,alcohol intake,fruit and vegetable intake were selected from the dietary adjustment household data set,with 11 consecutive variables.The indexes selected from the two data sets do not follow normal distribution.(3)the continuous bayesian network established by IPCB algorithm on two data sets is more complete than the discrete bayesian network established by MMHC algorithm,and more indicators related to uric acid can be found.In the data set of metabolic indicators,the discrete bayesian network established by MMHC algorithm found 9 edges,among which only the direct relationship between triglycerides and uric acid was found.The continuous bayesian network established by IPCB algorithm found 13 edges,among which the relationship between age,triglyceride,HDL,LDL and uric acid was found.In the diet adjustment household data set,a total of 9 edges were learned from the discrete bayesian network,among which only BMI was directly related to uric acid,but no relationship between dietary habits and uric acid was found.Continuous bayesian network learning to 14 edges found direct associations between uric acid and BMI and intake of meat,cooking oil,edible salt and fruits and vegetables.Conclusion:(1)simulation experiments show that IPCB algorithm can make full use of the information provided by the data,and the performance of building the continuous bayesian network framework is better than MMHC algorithm.(2)the bayesian network of uric acid-related factors established by IPCB algorithm found that the indicators directly related to uric acid were age,triglyceride,LDL cholesterol,HDL cholesterol,BMI,meat consumption,fruit and vegetable intake,edible salt and edible oil intake.Compared with MMHC discrete bayesian network structure learning algorithm,IPCB algorithm learns more complete network relations.(3)IPCB algorithm can effectively deal with continuous variables and obtain satisfactory results in the study of uric acid-related factors,which can provide new ideas for the study of chronic diseases related factors.