Research On Classification And Prediction Of Diabetes Based On Ensemble Learning

The international research results of diabetes mellitus demonstrate that the incidence rate of diabetes mellitus has been rising in recent years.With the continuous development of artificial intelligence and the acceleration of the digital process of medical care,it is of great practical significance to study the pathogenesis and early prevention of diabetes mellitus based on ensemble learning technology.On the basis of the research results of diabetes mellitus prediction at home and abroad,the relevant models are constructed: the relationship between various indexes in electronic medical record and blood glucose level is analyzed by means of statistics and ensemble learning;the value of blood glucose is predicted by use of ensemble learning;the ensemble learning with optimization algorithm is utilized to classify blood glucose.The specific research contents are as follows:1)With regard to the complex pathogenesis of diabetes mellitus and other problems,a blood glucose level correlation analysis model is proposed.Such model includes Pearson correlation test and feature importance evaluation of ensemble learning.Pearson correlation test is applied to analyze the significant correlation between each index and blood glucose value,while the feature importance evaluation of ensemble learning is employed to analyze the importance of each index to blood glucose value.The results suggest that the blood glucose level is affected by multiple organs and closely related to various indexes.2)As to the low prediction accuracy of single model,a prediction model of blood glucose based on Stacking fusion is put forward.The Boosting-series integrated model and SVR are selected as the first layer of blood glucose prediction model and linear regression as the second layer thereof.For the base model in the first layer,the hyperparameters are selected by use of random search and grid search.The results show that the prediction error of Stacking fusion model of hyper-parameter selection is lower and better than other advanced models,whose mean square error and explained variance are0.3774 and 0.1311 respectively.3)Aiming at the problems of ensemble learning model such as large number of hyper-parameters and complicated parameter adjustment,a GBDT classification model with wolf optimization is presented.This model makes classification relying on the common symptom data of patients with diabetes mellitus provided by UCI platform,and leverages the wolf optimization algorithm to automatically select the hyper-parameters in the ensemble learning model.The results indicate that the classification model through wolf optimization provides an accuracy of 0.9941,which is better than other parameter adjustment methods,and further manifest that the early symptoms of patients may be used as the main basis and reference for the early prevention of diabetes mellitus.Figure 26;Table 11;Reference 67...