A Data Driven Approach to Identify the Pseudo-periodic Behavior of Blood Glucose in Diabetic Patients

In this work we present a data driven approach to identify the pseudoperiodic dynamical structure in blood glucose in diabetic patients. In this work we use the pre-breakfast, pre-lunch and pre-supper blood glucose measurements of 70 patients over an extended period of time. We employ an algorithm that automatically learns the inner dynamics of the given data. This technique recursively discovers the underlying structure and is updated as new data emerges using radial basis functions. This method is very fast and memory efficient. Our modeling technique has done well to demonstrate the covering curves for the pseudo-periodic curves of blood glucose. The geometry of the covering curves identifies hyperglycaemia and hypoglycaemia in diabetic patients