TIME SERIES PREDICTION ARMA MODEL FOR PREDICTING BLOOD GLUCOSE IN ARTIFICIAL PANCREAS
1CIFHA CRECIL DIAS, SUREKHA KAMATH, SUDHA VIDYASAGAR
Patients with diabetes requires continuous monitoring of blood glucose level. Over the past few decades, Continuous glucose monitoring (CGM) has become a very helpful tool to manage and record glucose levels in the blood. With the help of CGM, the control and regulation of blood glucose can be achieved. Collecting the data from CGM, the paper attempts to predict future glucose levels by applying Auto Regressive Moving Average (ARMA) model. This predicted glucose level can be used for forecasting, and immediate appropriate action can be employed to avoid the risks related to diabetes. A good efficient model and a controller can be developed to improve the control using the Model Predictive Control technique with the predicted data set. The major risks that can be avoided are hyperglycemia and hypoglycemia. This paper elaborates on the estimation and prediction of blood glucose levels using the ARMA model in MATLAB platform. The CGM data is collected from a type 1 diabetic patient, and five-day data is recorded using the CGM device. The time series of the collected raw data is analyzed, and the parameter estimation is obtained. The model order is selected, and forecasting models are determined. This type of method for prediction gives good prediction with a lesser error when compared with original raw data and estimated values.
ARMA model, Diabetes, Prediction, Model estimation and Control