Research On Machine Learning Technology For Blood Glucose Prediction

Nowadays,increasing numbers of people are suffering from diabetes mellitus,but most of them are hard to know they have the disease.Diabetes is a very terrible chronic disease,which not only brings serious impact on the patient's own body and family life,but also puts tremendous pressure on the whole society's disease control management.Appropriate management of patients at risk with lifestyle changes and medications can decrease the risk of developing diabetes by 30% to 60%.Early detection of patients with elevated risk of developing diabetes mellitus is critical to the improved prevention and overall clinical management of these patients.In order to improve the awareness rate of patients with diabetes and thus manage the condition in a timely manner,this paper starts from the patient's metabolic data,analyzes in depth the influence of different examination items on blood glucose,and makes full use of the relationship between the examination subjects and the items.In addition,by the combination of prediction models,this paper takes full advantage of the different models' characteristics to improve the stability of the results.The main work of this paper is as follows.(1)Aiming at the complicated features of the patient examination items,this paper proposes a hierarchical feature selection method based on the sequence backward selection.First,due to the huge computational overhead and flexibility of the SBS algorithm,We perform the importance ranking of all features,and test the model effects of the remaining subsets in a targeted manner according to the ranking results to select the optimal subset.In addition,the data in the medical field has strong professionalism.In the feature selection process,only the examination items are generally considered,and the correspondence between the examination subjects and the examination items is neglected.This paper proposes a hierarchical feature selection method based on the hierarchical relationship between the examination subjects and the items.(2)Based on the characteristics of minority class distribution in unbalanced data sets,this paper proposes an unbalanced data processing method based on Region-SMOTE.In order to combine the boundary distribution information,the vote rule is formulated,which divides the minority samples into safety samples,intermediate samples and dangerous samples,and adopts different processing strategies for different samples.By labeling the blood glucose values of patients,the numerical prediction problem is transformed into a binary classification problem,and the minority class samples would be over-sampled to improve the overall classification performance.At the same time,another three data sets are extracted from UCI to verify the method.This paper studies the blood glucose prediction method based on machine learning to find and solve many problems existing in the field of medical prediction today from the perspective of practical application.We carry out our research from several aspects,including analysis and processing of raw medical data,the hierarchical feature selection method based on SBS,and the model combination prediction method.The good experimental results prove the effectiveness of the proposed methods.This work has guiding significance and practical value for making full use of medical data and establishing a standardized blood glucose prediction system.