The Modified ARIMA Predicting Algorithm Apply on Glucose Values Prediction

In order to predict the blood glucose values of diabetic patients, this study uses two Autoregressive Integrated Moving Average Model (ARIMA) models—the self-optimized ARIMA model and the ARIMA model based on Bayesian optimization to analyze historical data from the continuous blood glucose monitoring system (CGM) and the equipment calibration value as the training data set to foretell a patient's blood sugar level in the future in order to prevent hypo- and hyperglycemic episodes. CGM data from 8 patients obtained by Suzhou Municipal Hospital in Jiangsu Province, China, was used to validate the data in this paper on the two models. Obtain and compare the minimum Mean Square Error (MSE) values of the prediction results of the two models at 15, 30, and 45 min. In order to improve the accuracy of the patient's blood glucose level prediction, the model using the smallest MSE value among the two ARIMA models was used as a method for predicting the patient's blood glucose level prediction. The ARIMA model based on Bayesian have a good performance with 0.330 MSE value in 15 min prediction than self-optimized ARIMA model which could be preciously predict the glucose value.