Blood Glucose Prediction Based on Kalman Filter with a Savitzky-Golay Filter

The Artificial Pancreas System (APS) manage diabetes patients by providing real-time glucose data. However, glucose prediction remains a difficult challenge due to the dynamic, nonlinear, time-varying and various external factors such as meals, exercise, and stress. Additionally, continuous glucose monitoring (CGM) data often contains significant sensor noise, which complicates accurate predictions. In this paper, we propose a glucose prediction model based on kalman filter with a savitzky-golay filter. The kalman filter is an effective method for state estimation and prediction over time, enabling real-time tracking of glucose variations in diabetic patients. The feasibility of this prediction was assessed using data from the UVA-PADOVA virtual simulator. As a result, the results demonstrated the potential of the kalman filter for prediction, and we found that the kalman filter with savitzky-golay filter improves performance. These results provide foundational data that can be utilized for subsequent comparisons with more sophisticated deep learning-based models.