A deep transfer learning model for personalized blood glucose prediction

Abstract

The prediction of blood glucose concentration by deep learning can preferably guide a direction in the treatment of diabetes patients. However, insufficient medical data of individual patients and pathological differences between patients affect the performance of deep learning neural network models. This paper uses the authentic blood glucose data of hospitalized diabetic patients to train a deep learning model, and then uses the data of individual patients to adjust the model through transfer learning. The results demonstrate that the deep transfer learning model has better prediction performance than the original deep learning model, support vector regression (SVR), and BP neural network.