Predicting Blood Glucose Levels with LMU Recurrent Neural Networks: A Novel Computational Model

Type 1 diabetes disrupts normal blood glucose regulation due to the destruction of insulin-producing cells, necessitating insulin therapy through injections or insulin pumps. Consumer devices can forecast blood glucose levels by leveraging data from blood glucose sensors and other sources. Such predictions are valuable for informing patients about their blood glucose trajectory and supporting various downstream applications. Numerous machine-learning models have been explored for blood glucose prediction. This study introduces a novel application of Legendre Memory Units for blood glucose prediction. Employing a multivariate time series, predictions are made with 30-minute and 60-minute horizons. The proposed model is comparable with state-of-the-art models on the OhioT1DM dataset, encompassing eight weeks of data from 12 distinct patients.