A Comprehensive Review of Machine Learning and Deep Learning Approaches for Diabetes Prediction: Bridging Accuracy and Interpretability Towards a Two-Stage Comparative Framework

— Early diagnosis of diabetes mellitus, a chronic metabolic condition that affects more than 537 million adults worldwide, is crucial to minimize severe complications. The use of machine learning and deep learning has brought about a significant change in the prediction of diabetes by using non-invasive clinical variables. This review provides a critical analysis of two key studies: A Comparative Analysis of Diabetes Prediction Using Machine Learning and Deep Learning Algorithms in Healthcare [1] and A Study on Non-invasive Diabetes Causing Variables and Their Covariance Relationship in Diabetes Prediction Using Machine Learning Algorithms [2]. This review synthesizes methodologies, results, and limitations, emphasizing the strengths of ML and DL models, pinpointing research shortcomings with a two-stage comparative model that maximizes predictability while preserving interpretability and reducing computational time. The outcomes will be used in future studies related to hybrid models and treatments that can be applied clinically