Prediction of Diabetes Mellitus based on Logistic Regression and Random Forest

Abstract

Diabetes mellitus is a metabolic disease characterized by hyperglycemia caused by impaired insulin secretion and various degrees of peripheral insulin resistance. However, the primary mechanism is still unclear. The study analyzed medical data and used machine learning to build two predictive models to predict a patient's risk of heart disease. The dataset consisted of 3,000 clinical cases and 9 independent variables, such as high blood pressure, heart disease, and glycosylated hemoglobin. The target variable was whether the patient was diagnosed with diabetes mellitus. During this study, the logistic regression model initially had an accuracy of about 96% for the training set and then the fitted model was tested and the accuracy remained good at 94%. The prediction results of this logistic regression model were superior. In order to better predict the occurrence of heart disease, a second set of prediction models was developed in this study using Random Forest, with an accuracy of 100% in the training set, and then the fitted model was tested and the accuracy remained excellent at 96%. The two models used in this study provide a foundation for future testing of diabetes mellitus predictor variables and the discovery of even better predictive models.