Study On Mining Medical Quality Data Based On Large Scale Bayesian Networks

The quality of medical care has a strong impact on the relationships between thedoctors and patients. Through medical quality-oriented data mining approach, lookingfor all the possible factors that affect the medical quality indicators can help hospitalsimprove the medical care quality, reduce medical care costs, improve the marketstandardization of medical care. As a widely used data mining technology in themedical industry, bayesian network is able to describe the hidden relationshipsbetween the attributes. Most researchers analyze the data using statistical methods,and these methods cannot tell more expression of the potential knowledge. Theexisted bayesian method can combine the prior knowledge. In our research, weimprove the bayesian structure learning method, and learning the model around themedical care indicators. The main work of the research are as follows:(1)Presenting a Bayesian Network structure learning algorithm-MedicalK2: Inthe structure learning period, the number of potential Bayesian network structure is. MedicalK2uses greedy algorithm and Bayes posterior score as its core framework.We divide the set of properties into k sub set: a)Use random sequence inside a subset;b)Use topological sequence between different subset.(2)Building the Bayesian network model around the medical care qualityindicators: At present, most research on medical home page was focus on calculationand can’t find the potential relationship between the attributes. The Bayesian networkgroup building is around the medical quality indicators put forward by hospital.Though modeling about the indicators, the Bayesian network structure can reflect thedependency relationship between the attributes, and also reflect the factors that maycause the changes of the medical care quality.(3)Logically analysis and discuss the difference between the structure of theBayesian network model: In order to evaluate the potential relationship in theBayesian network model, we propose a method to analysis and evaluation of theBayesian network model based on Markov blanket: Comparative analysis theBayesian network model built from different data sets. Comparative analysis thedifference to identify a series of factors that cause the changing of the indicators. Based on the above, the research was about the dataset from three hospitals’ sixyear medical record. It’s proved the feasibility and effectiveness of the solution.