Machine Learning-Based Glucose Monitoring Techniques for Diabetes Management: A Review

The prevalence of diabetes mellitus has been a significant challenge for researchers as there is currently no known cure for this chronic condition. Hence, regular glucose monitoring is essential for diabetic patients to prevent serious complications. With the advancement of AI technology, the feasibility of ML in diabetes management, specifically in blood glucose monitoring, has become a prominent area of investigation for many researchers. Therefore, this paper comprehensively reviews different ML models, such as CNN, RNN, DT, SVM, and ARIMA, in the context of glucose monitoring. The review suggests that ML models can forecast future glucose levels within a certain prediction horizon while yielding reasonable accuracy. Such technology can ultimately improve glycemic control and prevent the serious complications of diabetes by taking appropriate treatments or actions in advance. However, challenges and limitations exist in the practical implementation of these models, including reliability, user-friendliness, and generalization. Future research should focus on developing universal and accurate models that can be tailored to specific patient groups while accounting for heterogeneity in glucose metabolism.