Abstract:
Accurately predicting blood glucose(BG) levels is crucial for effective management of type 1 diabetes. This paper proposes a hybrid neural network model for blood glucose...Show MoreMetadata
Abstract:
Accurately predicting blood glucose(BG) levels is crucial for effective management of type 1 diabetes. This paper proposes a hybrid neural network model for blood glucose prediction algorithm, which combines convolutional neural network (CNN) and Transformer. The algorithm takes into account various parameters, including patients’ historical blood glucose data, carbohydrate intake, and insulin injection amount, and integrates CNN and Transformer networks for comprehensive analysis. Specifically, the CNN is employed to extract local features from the input data, while the Transformer network captures global dependencies and contextual information in the time series data. The experiments were conducted using data generated by the UVA/Padova simulator, simulating blood glucose data for 10 adult individuals. Evaluation results demonstrate that the algorithm achieves average MAPE values of 2.41%, 2.89%, and 3.19% for the 15minute, 30-minute, and 60-minute predictions, respectively, with average RMSE values of 6.75, 10.51, and 15.98, indicating low prediction errors and good accuracy.
Published in: 2023 IEEE 9th International Conference on Cloud Computing and Intelligent Systems (CCIS)
Date of Conference: 12-13 August 2023
Date Added to IEEE Xplore: 02 October 2023
ISBN Information: