Abstract:

The goal of type 1 diabetes treatment is to achieve optimal and long-term control of blood glucose levels (BGLs). The automated prediction of BGL using machine learning (ML) techniques is seen as a potential method that may help achieve this goal. In this context, this work presents novel advanced ML architectures based on deep learning and ensemble learning to forecast BGL. The deepensemble models are created using unique meta-learning methodologies that test the viability of modifying the dimension of a univariate time series forecasting job. The models are tested both statistically and clinically. The suggested ensemble models' performance is compared to non-ensemble benchmark models. The findings demonstrate that the generated ensemble models outperform the developed non-ensemble benchmark models, as well as the usefulness of the suggested metalearning methodologies.

Description:

Blood glucose level, deep learning, diabetes mellitus, meta-learning, ensemble learning, time series forecasting