Predicting Short-Term Changing Blood Glucose Level of Diabetes Patients using Noninvasive Data

Diabetes is a chronic disease in which the blood glucose level stays high over a prolonged period. Uncontrolled diabetes put patients into critical conditions when the blood glucose level moves outside the clinically accepted range. Knowing the trend of the blood glucose level changes in advance will give the ability to take corrective actions to maintain it within the recommended levels. Continuous glucose monitoring (CGM) devices with an associated prediction model integrated into insulin pumps implanted in type 1 diabetic patients are able to predict the future glucose level in real time. A few different techniques for short-time glucose prediction have been reported incorporating past CGM readings, insulin dosage and meals. However, future glucose level predictions for patients without implanted insulin pumps are not available. This paper presents a machine learning algorithm to implement a short-term prediction model using non-invasive data; a past glucose level (measured or previously predicted), drugs dosage, food consumption, and life styles of diabetes patients. A deep learning time series forecasting based Long short-term memory (LSTM) model is used to predict the blood glucose level 30 minutes in the future. The model used data from diabetic patients to predict the short-term blood glucose levels with an accuracy level of 79.97%. The proposed model was able to predict blood glucose levels 30-min in advance with an average daily glucose levels error of 22.2 mg/dl.