Application Research Of Fusion Model Based On Ensemble Learning In Blood Glucose Prediction

With the increasing social pressure,people’s work and rest time become more and more irregular,many people’s eating habits are very unhealthy,which leads to the increasing number of undiagnosed patients with diabetes,and brings great hidden danger to people’s health.With the current medical conditions can not cure diabetes,only early detection and active prevention can slow down the harm of diabetes,but it takes a lot of manpower and material resources to know the blood glucose value of physical examination personnel in batches.If we can reasonably use the machine learning ensemble learning method to build the blood glucose value prediction model and accurately predict the blood glucose value,the medical staff can give early warning or treatment to those whose blood glucose value is higher than the normal level,so as to reduce the risk of diabetes and save a lot of medical expenses.In the study of blood glucose prediction based on physiological model,it is easy to be affected by individual differences and external factors.With the rapid development of machine learning,data-driven blood glucose prediction method highlights the advantages of high accuracy.Based on the ensemble learning model in machine learning,this paper uses the idea of integrating LightGBM model,XGBoost model,Cat Boost model,GBDT model,Linear Regression model and Stacking model,and combined with the relevant theoretical knowledge of machine learning,the corresponding improvement and optimization of the model are made,and a new model of blood glucose value prediction is constructed.Finally,the prediction results are compared and analyzed.The main research work of this paper is as follows:(1)Preprocess the data set.Because of the high dimension of the data set,and there are many abnormal data in the data set,it is very important to preprocess the data set before building the model.In this paper,we first deal with the missing values and outliers in the data set,encode the data set with One-Hot,then analyze the weight of features,and finally divide the data set into training set and training set.(2)The LightGBM model with optimized parameters is constructed.The LightGBM model optimized by three parameter optimization algorithms is proposed,namely HY--LightGBM model(optimized by Bayesian super parameter optimization algorithm),GALightGBM model(optimized by Genetic algorithm)and RS-LightGBM model(optimized by Random Search algorithm)to predict the blood glucose value.Finally,the prediction results were evaluated according to the mean square error(MSE)and other evaluation indexes.The experimental results show that LightGBM model has obvious advantages over XGBoost model,Cat Boost model,GBDT model and Linear Regression model without parameter optimization.The prediction accuracy of HY-LightGBM model optimized by parameters is better than that of LightGBM model and other regression prediction models optimized by other parameter optimization algorithms.(3)The blood glucose prediction model based on Stacking model fusion idea was constructed.The HY-LightGBM model,RS-XGBoost model and RS-Cat Boost model after parameter optimization are used as the base learners of the first layer of the fusion model,and the GBDT model and linear regression model are used as the meta learners of the second layer of the fusion model to form a fusion model to improve the prediction accuracy.The experimental results show that the fusion model based on Stacking model further improves the prediction accuracy of blood glucose value,and has good fitting ability.