Research On Diagnosis Of Heart Disease Based On Data Mining

In the past ten years,in the world,the main cause of death in heart disease belongs to the forefront,The total number of sudden cardiac death in our country is the largest in the world,the disease is considered to be the major illness in the middle and old ages.In the method of diagnosis of heart disease,angiography is the best method,but this method requires not only a high level of the hospital to detect.So the patients which live in areas with poor medical care can not get timely diagnosis and treatment.At present,our country is using some communication tool allows patients to communicate to the doctor for patients with heart disease treatment,and more recently,such as through APP makes patients and doctors to communicate online,However,these methods still exist the coverage and efficiency is not high enough.Therefore,the researchers through data mining actively looking for alternative methods of artificial diagnosis,and Especially in recent years,the prevalence of"big data" concept,many companies and researchers use the method of data mining in various fields,and the development of medical data for medical information is of great significance,so in this paper,we propose a method of using data mining to diagnose heart disease.The most important step in data mining is the modeling process,because of the deep learning than shallow learning has a better ability to express data.In this paper,we use a method of deep learning,deep belief network model(DBN),as data pro-cessing model in data mining.Firstly,the data of patients with heart disease were pretreated,then through the DBN model of the patient's data model to prediction results.Although DBN has its advantages compared with shallow learning,but there are still some space for improvement after DBN has been put forward:(1)From the point of view of DBN model,Aiming at the difficult problem of DBN network layer,From the perspective of information theory,by analyzing the physical meaning of information entropy,after the training restricted Boltzmann machine(RBM)system has reached a stable state,Calculating the information entropy of hidden layer,By comparing the information entropy of each layer,the highest level of information entropy is selected as the last selected DBN layer.(2)Based on the data set of deep learning,the training of DBN requires a large number of data sets,but if the data is relatively small,there will be a problem of overfitting,In this paper,we use the Fisher criterion and the regularization method to adjust the weights of DBN training,so as to prevent the problem of inaccurate prediction.The experimental part of the DBN network layer and the data set for the results are discussed,and the improved DBN network prediction results were compared with the BP neural network,the results show that the improved DBN neural network can diagnose heart disease better to complete the work.