Prediction Of 30-day Readmission In Diabetic Patients Based On Neural Network

With the change of social environment and life style,more and more urban and rural residents are suffering from chronic diseases.As a progressive chronic disease,diabetes does not seriously endanger life or cause pain at the beginning.Instead,it hibernates in the human body,affecting other organs and systems step by step,causing various complications and affecting physical health and daily life.People also began to pay more attention to the prevention and treatment of diabetes,among which the problem of readmission is also a manifestation of the treatment effect and severity of patients.Short-term readmission may be caused by poor drug treatment and ineffective control of diabetes,or by failure to grasp the development trend of the patient’s condition,leading to diabetes recurrence.Therefore,readmission prediction of patients can assist doctors to provide more reasonable plans for followup treatment of patients.In the era of information technology development,first of all,medical electronic records provide a large number of different forms of data,and readmission prediction is made based on the available information mined from them,which not only finds more valuable data,but also improves the reliability of prediction.Secondly,The application of machine learning model in various fields and its continuous development have had a great impact on social operation.It is an effective way to achieve the value of data to find the rules of data through model learning and apply it to the prediction of new data.The models used for different problems are also different,and the model based on diabetic readmission prediction also needs to be constantly improved to try different algorithms and parameter optimization methods to achieve the best performance.Therefore,the classification and prediction of readmission from the two aspects of data and model not only makes efficient use of data,but also improves learning efficiency and accuracy of the model.In this paper,the relationship between characteristics and readmission was analyzed by studying the medical records of patients with diabetes.In data value processing,missing value and outlier value are analyzed and treated differently according to the situation.In feature processing,new features are constructed by combining attributes,and three feature selection methods are integrated to retain important features.In order to avoid the bias of model training due to the unbalanced proportion of positive and negative samples,this paper adopts the method of comprehensive sampling to balance data categories.Finally,based on attention mechanism,this paper builds the coding network model,used to play a role of feature extraction,and then the encoding model of the self encoding network is combined with the full connection to build a classification model to predict the readmission of diabetes.By comparing the various experimental indicators with the three models of Support Vector Machine,Decision Tree and Random Forest,it is shown that the neural network model built in this paper performs better.