Using machine learning for predicting the risk of glucose level drop during exercise using continuous glucose monitoring data

Abstract

Managing blood glucose during intense physical activity remains a critical challenge for individuals with diabetes. This thesis presents a data-driven approach for predicting whether a planned exercise session is likely to lead to a high glucose drop. The prediction is based not only on session characteristics such as duration and intensity but also on contextual and individual-level factors including past activity patterns, continuous glucose monitoring (CGM) data, and patient-specific information such as age and health status. The goal is to support proactive self-management by providing personalized alerts and recommendations before exercise begins.