Blood Glucose Concentration Prediction Based on Canonical Correlation Analysis

Diabetes is a chronic disease that requires diabetics to monitor their blood glucose levels to maintain them within normal limits. Therefore, prediction of blood glucose concentrations has been a field of great interest in clinical research. Compared with many data-driven modeling studies, using the patient's historical blood glucose value, food intake, insulin injection volume, and exercise volume as input to the prediction model, this paper only uses the historical blood glucose value as input to make subsequent predictions, thus reducing the measurement error that exists in the input data portion. In order to predict blood glucose, this paper uses a paradigm correlation analysis. Although this method has been widely used, it has not been involved in blood glucose prediction. Therefore, for the first time, this article uses a canonical correlation analysis to predict blood glucose in people with type 1 diabetes. The true measurement data of10 type 1 diabetics was used to verify the prediction effect. The input of the experiment selects the historical data of 20 min for each patient, and the predicted horizon is 5, 10, 15, and 20 min respectively. The average values of the corresponding root mean square errors are: 6.096, 12.022, 17.384, 21.713 mg/dl. The results of this study are compared with previous prediction methods, indicating that the canonical correlation analysis method has application prospects in blood glucose prediction and achieves high-precision prediction.