Machine Learning and Deep Learning Models for Predicting the Onset of Diabetes: A Pilot Study

Abstract

Diabetes currently one of the most significant worldwide concerns, and its prevalence is only expected to increase in the future years. In order to monitor glucose levels in the blood and set treatment protocols for diabetes, keeping a regular schedule for checking blood glucose levels is essential. The purpose of widespread adoption of digital health in recent years has been to enhance diabetic healthcare for patients, and as a result, a massive quantity of data has been collected that may be used in the ongoing management of this chronic condition. Deep learning, a relatively new kind of machine learning, is one method that has taken advantage of this trend, and its applications seem promising. In this research, we provide a thorough analysis of how deep learning has been used in the study of diabetes thus far. We conducted a comprehensive literature search and found that this method is most often used in the following settings: diabetes diagnosis, glucose control, and the identification of diabetes-related complications. We have described the most important details regarding the learning models used, the development process, the primary outcomes, and the baseline techniques for performance assessment from the 40 original research publications that we selected based on the search. In the reviewed literature, it becomes clear that several deep learning algorithms and frameworks have outperformed traditional machine learning methods to attain state-of-the-art performance on numerous problems involving diabetes. However, we point out several gaps in the existing literature, such as a dearth of readily available data and a lack of clarity in the interpretation of models. In the near future, these obstacles may be surmounted thanks to the fast advancements in deep learning methodologies which will allow for wider application of this technology in therapeutic settings.