Precision Insulin Delivery: Predictive Modelling for Bolus Insulin Injection in Real-Time

Abstract

Insulin is recommended for patients with Diabetes Mellitus (DM). It is challenging for doctors to prescribe accurate bolus insulin before every meal due to real-time factors such as the size of the meal, skipping a previous meal, and physical activity, which can risk the patient towards hyperglycemia or hypoglycemia. Previous studies executed insulin predictions where the methods did not consider the cases of controlled glucose levels, type of insulin prescribed, time of insulin-induced, and data detersion that can alter the predictions. To address these problems, our work has proposed an insulin predictive model from the integration of Internet of Things (IoT) devices, i.e., Continuous glucose monitoring (CGM) sensor and insulin pumps with rapid-acting insulin type where the insulin dosage with corresponding Current Blood Glucose levels (CBG) and improved Next Blood Glucose levels (NBG) are chosen. The dataset is subjected to data detersion where pre-processing, Exploratory Data Analysis (EDA), and Feature Selection is performed. Machine Learning (ML) models are applied on curated dataset where Decision Tree (DT)-Bagging algorithm, performed the best with a Mean Absolute Error (MAE) of 1.54 and a Mean Square Error (MSE) of 4.15. Performance metrics of the current study imply its suitability in medical applications for accurate prediction of real-time insulin dosage.