Scoping Review on Deep Learning Model for Classification and Prediction of Diabetes

Background Diabetes mellitus is a growing global health challenge, particularly in low- and middle-income countries, where delayed diagnosis and inadequate treatment contribute significantly to complications and mortality. Artificial intelligence, particularly deep learning (DL), has emerged as a promising tool for improving diabetes classification and outcome prediction. Objective This scoping review aimed to map existing evidence on the development and application of one-stage and two-stage deep learning models for the classification and prediction of Type 1 and Type 2 diabetes. Methods A scoping review was conducted using PubMed and Google Scholar databases, guided by the Population-Concept-Context (PCC) framework and PRISMA-ScR methodology. Studies were included if they applied deep learning models to the classification and/or prediction of diabetes. Data extraction was performed using a structured spreadsheet capturing model type, dataset, features, and performance metrics. Results Out of 750 identified studies, 50 met the inclusion criteria. Convolutional neural network-based architectures were the most common (16; 38%), followed by recurrent neural networks and hybrid models. The majority of studies (43; 86%) used a one-stage deep learning approach integrating classification and prediction into a single step. Only 7 studies (14%) employed a two-stage framework, and none were conducted in the African context. Common datasets included the Pima Indian dataset and the UCI Machine Learning Repository, with limited use of local or clinical datasets. Conclusion Deep learning models demonstrate strong potential for improving diabetes diagnosis and prediction. However, the dominance of one-stage models and the lack of African-based studies highlight critical methodological and geographical gaps. Future research should explore two-stage models tailored to local datasets to enhance clinical relevance and promote global equity in AI-based diabetes care.