Classification of Personalized Daily Data from Continuous Glucose Monitoring

During the last few decades, the number of people suffering from diabetes has been increasing. The artificial pancreas (AP) is a closed-loop glucose controller that successfully controls blood glucose (BG) levels, provide automatic insulin delivery, and reduce the risk of hypoglycemia. While recently proposed methods for the AP have shown good results, there many challenges remain in designing successful algorithm. Precise and continuous monitoring of BG levels is required for effective treatment of patients. In this paper, we present a comparison of daily continuous glucose monitoring (CGM) time series data and perform classification for automated control systems. A training dataset of six different patients with type 1 diabetes (T1D) was used to analyze their daily CGM profiles. We compared daily CGM profiles to calculate their minimum value using the root-mean-square error (RMSE) and Manhattan distance. A correlation was made between similar daily CGM profiles using the unweighted pair-group method with arithmetic mean (UPGMA) hierarchical clustering method. The results indicated that the proposed clustering method forms groups with similar values thus indicating various stages of patient's condition.