A Diagnostic Assistant Model For Diabetes Complications Based On Hospital Information System

Currently,the worldw ide diabetic population in the 20-79 age group is about 425 million,and about 4 million of the diabetics die every year,diabetes accounts for 10.7% of all deaths in the world.Diabetic comp lications are the main factors affecting the life’s qualit y of diabetics and leading to their premature death,therefore,the health manage ment of diabetes and its comp lications is of great significance.This paper aims to assist phys icians in the diagnosis of diabetic complications to make up for the limitation of phys ic ian’s diagnosis and avoid the occurrence of misdiagnosis of diabetic comp lications from the perspective of establishing a diagnostic assistant model for diabetic complications.It focuses on the proble ms and limitations of the related research about diagnostic model,diagnosis of diabetic complications and the multi-label classification methods,and tries to impove them.All the work of the paper contr ibutes to the prevention and treatment of diabetes.Focusing on the establishment of a reasonable and well-diagnosed mult i-label classification model for diabetic complications,the following work is ma inly done:(a)C oncerns about the correlation between the diabetic comp lications in the promotion of model diagnostic performance,associat ion rules and χ2 independence test are used to analyze the correlation;(b)Univariate logist ic regression analys is and SVM-RFE method are applied to feature selection of each diagnostic model for diabetic peripheral neuropathy,diabetic peripheral vascular disease,and diabetic nephropathy;(c)six machine learning methods are used to establish a diagnostic model for each of the three diabetes complications,and the optimal hyperparameters of each model are explored,then the three diagnostic models for each of the diabetic complications are compared and analyzed.(d)The chain order and the base classifier of the classifier chain model are improved to fit the classification proble m in the context of this study,and the binary relevence and the improved classifier chain methods are used to establish a mult i-label classification models for diabetic complications,the two models’ performances are compared and analyzed.The results of the paper are that:(a)there are certain correlat ions between the patients’ diseases,especia lly the diabetic complications;(b)random forest has the best diagnostic classification performance for diabetic peripheral neuropathy,Ada Boost has the best diagnostic classification performance for diabetic peripheral vascular disease and SVM has the best diagnostic classification performance for diabetic nephropathy.(c)the mult i-label diagnostic model for the three diabetic complications based on the improved classifier chain method is superior to the model based on binary relevence method from the mult i-label classification evaluation metrics,and from the single-label classification evaluation metrics,the diagnosis model based on the classifier chain method has a significant performance improvement for the diabetic peripheral vascular disease.The results confir m the rationality,applicability and validit y of the improved classifier chain in this research issue.