Machine Learning and Explainable Artificial Intelligence for Diabetes Prediction: A Systematic Review with Bibliometric Insights

The use of Machine Learning (ML) has emerged as an effective approach for the early prediction of Diabetes mellitus, thus enabling early diagnosis and personalised healthcare. However, the increasing complexity of ML models have raised concerns regarding transparency, interpretability and clinical trust which has led to the adoption of Explainable Artificial Intelligence (XAI). In this review, the authors provide a comprehensive summary of the relevant recent research on ML and XAI models for diabetes prediction. They systematically analyze 30 peer-reviewed papers published in the last few years. Selected literature is analyzed for algorithms, datasets, prediction performance, methods, explainability technique and trends in the research. Bibliometric and comparative analyses are conducted in order to provide a comprehensive overview of the research landscape, which includes publication trends, frequency of algorithm usage, distribution of datasets, classification of methodologies, ML–XAI heatmaps, and visualization of keyword co-occurrence and network. The results show that the ensemble learning methods such as Random Forest and XGBoost outperform each other in predictive performance. SHAP and LIME are the most popular XAI techniques for improving model interpretability. Great progress has been made, but challenges remain in the areas of heterogeneous clinical data handling, model generalizability, interpretability and model deployment in real-time in a healthcare setting. Considering the research gaps identified, suggestions for future research directions are proposed in the scope of hybrid explainable frameworks, multimodal healthcare data integration, federated learning, and clinically trustworthy AI systems for accurate and transparent diabetes prediction.