Controlling blood glucose levels in diabetics by neural network predictor

In this study we develop a system that uses some variables such as, level of exercise, stress, food intake, injected insulin and blood glucose level in previous intervals, as input and accurately predicts the blood glucose level in the next interval. The system is split up to make separate prediction of blood glucose level in the morning, afternoon, evening and night, using data from one patient covering a period of 77 days. We have used RBF neural network, and compared our result with MLP neural network that was implemented by the others. The assessment of the analysis resulted in a Root Mean Square Error of (0.04 +/- 0.0004) mmol/l.