A Particle Swarm Optimization Based Nonlinear Autoregressive Model for Predicting Plasma Glucose Level following Ivgtt

An important aspect of real time glucose control in diabetic subjects is forecasting the future glucose profile. From a modeling perspective, the complexity of interactions between plasma glucose and its controlling hormone insulin, demands a nonlinear modeling approach for representing the dynamics of glucose metabolism. In this paper, a technique based on the hybrid approach of nonlinear autoregressive modeling and particle swarm optimization (PSO) has been applied for modeling glucose metabolism and predicting the glucose profile from past glucose-insulin time series data. The identification of nonlinear autoregressive model involves selection of proper terms from a family of regressors, so as to minimize the deviation between the model and actual output. However, since the existing selection approaches do not involve the reduction of overall error, but it's minimization, the final model obtained is not always globally optimal. In this regard, the identification process has been framed as an optimization problem and solved using PSO. The proposed model has been analyzed on the clinical data of sixteen subjects. It has been demonstrated how the combination of an evolutionary computing technique and nonlinear autoregressive modeling is more efficient in predicting the evolution of glucose profile as compared to the widely used minimal model.