Multivariate Modelling for Prediction of Time-Series Blood Glucose Level using Vector Autoregression (VAR)

Abstract

Predicting glucose levels remains a significant challenge in diabetes management, with various factors influencing regulation. Modern technologies like AI and ML offer potential solutions by implementing prediction systems. This study focuses on utilising the Vector Autoregression (VAR) method to make accurate predictions, considering factors such as insulin and meal intake. Ten datasets, including blood glucose levels, carbohydrate intake, and insulin intake, were collected using MATLAB Simulink simulations. Python was then used to build predictive models with a 70:30 and 80:20 ratio for training and testing. The VAR model's prediction performance was evaluated using metrics like MAE, RMSE, and MSE. The 80:20 data split with binary insulin values yielded better results for blood glucose prediction, with MAE of 11.25808, RMSE of 12.36846, and MSE of 184.3054. This study offers insights into time series prediction of blood glucose using the VAR machine learning model, potentially enhancing diabetes care.