Prediction of Blood Glucose Levels and Diabetes Complications Using Wearable’s for Type1 Diabetics

Diabetes leads to early passing and impairment worldwide and affects people despite nation, age and sex. Accurate and timely prediction of blood glucose levels is important for managing diabetes, enabling proactive interventions to prevent hypo- and hyperglycemic events. Though a prediction of glucose level is a critical aspect in the real world, it helps diabetic patients manage their conditions and lowers the risk of developing health complications. This study aims to predict blood glucose values of diabetic patients using LSTM and GRU individual models on D1NAMO dataset. Forecasting of glucose level was carried out by considering diabetic patients CGM recordings additionally with their physical measurements. The comparison of these models for blood glucose forecasting under three evaluation metrics was performed. We successfully implemented our proposed approach for nine type1 diabetic patients and achieved RMSE 0.165, MAE 0.133 and 1.605 MAPE. It has been found that a single layer of GRU with a dense layer is sufficient to obtain good accuracy. Additionally, this paper includes predicting diabetes symptoms based on six glycemic diabetes ranges using SVM.