A Comprehensive Review on Prediction of Blood Glucose Level in Type 1 Diabetic Using Machine Learning Techniques

Abstract

Recent technological advancements have opened the doors for type 1 diabetes management systems, such as continuous glucose monitoring (CGM), advance sensors, artificial intelligence, and other hybrid models. Modern diabetes management systems provide accurate prediction of blood glucose level. In the last decade, researchers have given more emphasis on development of artificial pancreases (close loop system) by implementing various machine learning techniques with conventional diabetes management system that can be used for controlling blood glucose level in normal range to prevent critical abnormalities in individuals like retinopathy, neuropathy, nephropathy, seizure, and others. Predictive machine learning models form the core of next generation artificial pancreases. These models can be applied to decision making processes for on-time delivery of exogenous insulin to T1DM patients in order to prevent hypoglycemia and hyperglycemia by forecasting blood glucose concentration in blood for longer prediction horizon. Performance of predictive models depends on a number of factors. This paper incorporates a comprehensive review and compact guide of various data-driven models and the methodologies presented in the recent times for prediction of blood glucose dynamics, considering various parameters like source of dataset, nature of the dataset, pre-processing method of raw data, various configuration of input signal along with CGM data, machine learning techniques, and performance metric.