Predição contínua do nível de glicose no sangue em pacientes com Diabetes Mellitus Tipo 1 utilizando técnicas de machine learning

Abstract

Blood glucose control poses a significant challenge for individuals living with Type 1 Diabetes Mellitus (T1DM). Maintaining blood glucose levels within the normal range is a complex task that requires lifelong commitment and management. Fortunately, advancements in technology such as continuous glucose monitoring, insulin pumps, and mobile applications have enabled the close monitoring of various indicators related to diabetes management. This has led to an increased utilization of machine learning algorithms to gain insights into blood glucose behavior. In this study, we investigate the effectiveness of two machine learning techniques, namely Multilayer Perceptron and Gradient Boosting, for predicting blood glucose levels. The evaluation is conducted using a dataset collected over a period of approximately eight weeks, involving real-life conditions and data from twelve different T1DM patients. The numerical results demonstrate the capability of both methods to provide patients with information regarding their future blood glucose levels, highlighting their potential in assisting individuals with T1DM in managing their condition.