Diabetes Mellitus Prediction: Systematic Review of Prediction Models

The need for better and more sophisticated diagnostic methods has increased due to the sharp rise in diabetes incidence globally. Over recent times, Machine Learning (ML) and Deep Learning (DL) approaches have been identified as potential solutions for developing intelligent systems capable of diagnosing diabetes effectively and accurately. A major problem that affects this area is that comprehensive datasets on diabetes are hard to collect since they are usually based on laboratory tests or invasive procedures only. Therefore, this proposed approach accentuates ways in which anthropometric indicators and other non-invasive measures can become cost-effective substitutes that still exhibit high accuracy. Numerous studies have supported models using non-invasive features to perform competitively. The review discusses famous databases, algorithms, feature selection approaches, and evaluation criteria, and compares current research work. Lastly, it focuses on current shortcomings as well as presents areas or issues that future studies may investigate; such topics include hybrid models, explainable AI, and broader usage of non-invasive diagnostic solutions.