Comparison of MLP and elman neural network for blood glucose level prediction in type 1 diabetics

One of the most dangerous symptoms of Type I diabetes is the frequent and grate oscillation of blood glucose level that can lead the patient to unconscious and coma states. So being able to predict and finally prevent these two symptoms would simplify the management of the diabetic patients. This paper attempts to comparison the performance of MLP and Elman neural networks to predict the blood glucose levels in typel diabetics. Data set, used in this paper consists of the protocol of a 10 Iranian typel Diabetic women and include features such as type and dosage of injected insulin, The period of time (in hour) between two consecutive measurements of the blood glucose level, carbohydrate intake, exercise and the blood glucose level measured at start of the given period of time. Finally we concluded that the usage of Recurrent Neural Network such as Elman can be an appropriate model to predict the long term blood glucose level in type I diabetics also we could successfully increase the accuracy of prediction and reduce the number of layers and neurons used in the construction of Neural Networks.