Transform Diabetes - Harnessing Transformer-Based Machine Learning and Layered Ensemble with Enhanced Training for Improved Glucose Prediction.

Abstract

Type 1 diabetes is a common chronic disease characterized by the body’s inability to regulate the blood glucose level, leading to severe health consequences if not handled manually. Accurate blood glucose level predictions can enable better disease management and inform subsequent treatment decisions. However, predicting future blood glucose levels is a complex problem due to the inherent complexity and variability of the human body. This thesis investigates using a Transformer model to outperform a state-of-the-art Convolutional Recurrent Neural Network model by forecasting blood glucose levels on the same dataset. The problem is structured, and the data is preprocessed as a multivariate multi-step time series. A unique Layered Ensemble technique that Enhances the Training of the final model is introduced. This technique manages missing data and counters potential issues from other techniques by employing both a Long Short-Term Memory model and a Transformer model together. The experimental results show that this novel ensemble technique reduces the root mean squared error by approximately 14.28% when predicting the blood glucose level 30 minutes in the future compared to the state-of-the-art model. This improvement highlights the potential of this approach to assist diabetes patients with effective disease management.