Blood Glucose Prediction for Type 1 Diabetes using Machine Learning Long Short-term Memory based models for blood glucose prediction

Abstract

In this thesis, walk forward testing is used to evaluate the performance of two long short-term memory (LSTM) models for predicting blood glucose values for patients with type 1 diabetes. The models are compared with a support vector regression (SVR) model as well as with an auto regressive integrated moving average (ARIMA) model, both of which have been used in related research within the area. The best performing long short-term model produces results equal to those of the SVR model and it outperforms the ARIMA model for all prediction horizons. In contrast to models in related research, this LSTM model also has the ability to assign a level of confidence to each prediction, adding an edge in practical usability.