Bayesian GRU를 이용한 확률론적 혈당 수치 예측 모델 개발

Abstract

Diabetes is a metabolic disorder caused by abnormal insulin secretion, where continuous and accurate glucose management is crucial for improving the quality of life for diabetic patients. The development of Continuous Glucose Monitoring (CGM) technology has enabled precise analysis of real-time glucose data, fostering ongoing research into glucose prediction. This study proposes a multi-layer Bayesian GRU (SBGRU) model, combining conventional multi layers GRU model with a Bayesian approach, to enhance the accuracy and reliability of glucose predictions. Modeling was conducted using 21 virtual patients’ data generated through the FDA-approved UVA/Padova simulator, and Bayesian techniques were employed to model the inherent uncertainties in the predictive model. Consequently, the SBGRU demonstrated superior performance compared to existing models, with an RMSE of 11.85 mg/dL, MARD of 8.50%, and an 𝑅2 coefficient of 0.93, particularly showing high accuracy and reliability in hypoglycemic events. The proposed model significantly improves the reliability of glucose predictions, and such enhancements are expected to be vitally beneficial for the glucose management of diabetes patients.