A Study On The Blood Glucose Value Prediction Models For Diabetes Mellitus

With the continuous development of China's economy,people's living standards have been generally improved,but people's awareness of healthy eating habits is still weak;resulting in the rapid growth of the number of diabetes which endangers people's health seriously.At present,there is no radical cure for diabetes.If the situation is not under control,diabetes is easy to cause various concurrent symptoms,only through early detection and early treatment can reduce its incidence.the early diagnosis and preventive measures against diabetes is still not perfect.Therefore,it is necessary to establish a blood glucose value prediction model for diabetes,and identify the high-risk group in advance,and assist the doctors to diagnose.This is of great significance for improving people's health and promoting social harmony and stability.Based on the medical examination data of the domestic top three hospitals,this article first explains the basic theories of support vector machines,ridge regression,random forests,and XGBoost;and then preprocesses the data,removes outliers and noise data,and fills in features with fewer missing values.Standardize the data of different dimensions;After the data preprocessing is completed,we do feature selection.In this paper,two integrated learning methods are used for feature selection to avoid the errors caused by single model feature selection,we select 17 most important features.then build models based on the relevant theoretical basis of Chapters 2.In the SVR(RBF)model,the mean square error(MSE)is 1.51,and the average absolute percentage error(MAPE)is 10.68%.In the kernel ridge regression(RBF)model,its MSE is 1.47,MAPE is 11.26%.In the random forest model,its MSE is 1.45,MAPE is 10.41%,in the XGBoost model,its MSE is 1.39,and MAPE is 10.31%.Comparing the above four models,it can be seen that the prediction effect of the XGBoost model is better than that of the random forest,SVR(RBF),and KRR(RBF);The random forest model runs slower as the base learner increases,while the XGBoost model runs faster.Finally,the XGBoost model is used as the best model in this diabetes blood glucose value prediction.