Research On Diabetes Prediction Model Based On Deep Belief Network

In recent years,the number of people with diabetes in China has become the world’s first.Diabetes is a chronic disease that seriously harms human health,which can lead to the occurrence of multiple complications,and has a high mortality rate.Significant threat.In addition,diabetes has the characteristics of being difficult to diagnose and hard to cure,and it causes great loss of medical resources every year.The routine diagnosis of diabetes is mainly based on the inspection data and the doctor’s clinical experience,which is prone to missed diagnosis and misdiagnosis,which leads to the inability of patients to be treated in time and increases the risk of complications.Deep learning is a new field developed by artificial neural networks.At present,deep neural network methods are widely used in speech recognition,image recognition and other fields.It has the advantages of high accuracy,strong expression ability,and good transferability.Therefore,it is necessary and meaningful to apply deep learning to medical treatment,and to apply a trained algorithm prediction model to diabetes prediction.Early detection,early diagnosis,and early treatment reduce the possibility of complications and mortality,and assist doctors in making more comprehensive and reliable diagnosis and treatment decisions.According to the differences in the living habits of people in different regions,different living habits will affect the quality of the model.Models designed in one area are usually not applicable to another area.For area A,the prediction of diabetes is a strong relevant factor.The attribute is irrelevant in area B.Therefore,this topic analyzes and mines the data collected locally to ensure that the model obtained is more meaningful to the local people.In addition,because traditional machine learning algorithms cannot effectively use high-dimensional attribute information,this topic studies the prediction model of diabetes based on deep trust network.First,analyze the advantages and disadvantages of traditional algorithms,perform data cleaning,and analyze the relationship between diabetes and its related factors.Sex.Then the common logistic regression method and support vector machine method are used to model the data set.At the same time,a suitable deep learning model is selected for training.Finally,based on the ROC curve and the confusion matrix method,the performance and prediction results of deep learning models and other machine learning algorithm models are compared.Experiments show that models based on deep trust networks are better than support vector machine models and logic in terms of accuracy and sensitivity.Regression model.Finally,in view of the poor stability of traditional deep trust networks,the network model is improved,and a multi-branch deep trust network model is proposed.Experiments show that the prediction effect of the multi-branch trust network has improved significantly.