Machine learning algorithms for blood glucose prediction in diabetes management

People with type 1 diabetes (T1D) depend on externally administered insulin to regulate their blood glucose levels and prevent long-term complications. Managing diabetes involves a com- plex set of tasks, including glucose monitoring, insulin administration, carbohydrate counting, and maintaining a regular exercise routine, among other lifestyle choices. Achieving the recommended glycemic target is challenging, with only about one out of four individuals successfully managing their condition. Clinical wearable devices, such as continuous glucose monitors, insulin pumps and activity trackers, play a crucial role in helping people with T1D to control their blood glucose levels. Artificial intelligence (AI) and decision support systems that leverage the wealth of data generated by these devices can offer significant assistance in managing T1D effectively and enhancing the overall quality of life of those living with the condition. This thesis focuses on building Machine Learning (ML) models using glucose, insulin, meal, and activity data, to predict potential future adverse events, such as nocturnal hypoglycaemia, recommend carbohydrate intake to prevent low glucose levels, or forecast forthcoming blood glucose values. These models can be used as the backbone of an AI decision support system for T1D management outside of clinical settings, mitigating the risk of health complications and the progressive deterioration of the disease.