Using WOA with Feed Forward Neural Network in Prediction of Subcutaneous Glucose Concentration for Type-1 Diabetic Patients

Abstract

AbstractInsulin therapy for type 1 diabetic patients often results in high variability in blood glucose concentration and may cause hyper-and hypoglycemia. Insulin is known with its time lag action, there is a time lag between insulin infusion and its effect on lowering blood glucose (BG) concentration. Designing a closed loop control system for blood glucose concentration needs a good performed predictor, especially for long prediction horizons. Neural networks are widely used in blood glucose prediction with good performance, especially for short prediction horizons. Improving prediction performance of feed forward neural network (FFNN) for higher prediction horizon values is an aim of many research topics. In this paper, we propose an algorithm that uses whale optimization algorithm (WOA) in training feed forward neural networks that are already used before in BG concentration prediction. The results show that, WOA in BG predictor training reduces RMSE and Normalized prediction error (NPE) for all prediction horizons (PHs). The performance enhancement increases with increase of prediction horizon. FIT value of the predicted BG after 60 min is also increased from 47.6% to 63.1% when WOA is added in FFNN training. The performance of our prediction model is comparable with other neural network referenced prediction models. We developed a FFNN predictor that is able to predict 60 min ahead.KeywordsFeed forward neural networkWhale optimization algorithmBlood glucose predictorPrediction horizon