Development of Generative-AI based Recommendation System for Diabetes Mellitus

Abstract

Diabetes Mellitus is a significant chronic physiological condition that requires early prediction, continuous monitoring, and personalized recommendations. Conventional approaches to diabetes prediction primarily focus on diagnosis and risk assessment, while providing limited support for generating context-aware management plans and patient-centric recommendations. This study proposes a Generative AI-based framework for the early prediction and intelligent management of Diabetes Mellitus by integrating risk stratification, Explainable Artificial Intelligence (XAI), Large Language Models (LLMs), and Retrieval-Augmented Generation (RAG). The proposed framework incorporates seven interconnected layers encompassing data acquisition, preprocessing, risk assessment, explainability, knowledge retrieval from trusted and reliable sources, and the generation of patient-centric recommendations, supported by a feedback-driven optimization mechanism. By integrating early diabetes prediction with explainability, evidence-grounded knowledge retrieval, and context-aware management plans and recommendations, the proposed unified AI-driven framework aims to address the limitations and research gaps associated with conventional prediction models. The effectiveness of the framework will be evaluated using quantitative performance measures, including accuracy, precision, recall, F1-score, sensitivity, and specificity, along with a comprehensive assessment of the applicability and explainability of the generated recommendations. The expected outcome of the unified framework is an adaptive, clinically supported, and personalized system that contributes to early diabetes prediction and effective disease management.