AI-Driven Personalized Meal Recommendation System for Diabetes Management

The effectiveness of diabetes management relies greatly on dietary discipline, yet most existing recommendation systems do not offer personalized meal guidance to diabetic patients. This research creates a machine learning–based recommendation system for the dietary needs of patients with Type 1 and Type 2 diabetes. Our system advances generic nutrition platforms through medical threshold inclusion, real-time personalization, and sophisticated classification approaches to offer precise individualized meal guidance. We removed all nutritionally inadequate and unfinished records from an initial database of 230 000 meals sourced from a publicly available Kaggle.com dataset, refining i t t o 1 3,000 v alidated r ecipes. F eature e ngineering focused on carbohydrates—breaking down sugars, calculating total carbohydrate amounts and sugar ratios, and estimating glycemic load. Five models—Long Short-Term Memory (LSTM), Random Forest, Artificial N eural N etwork (ANN), N aïve B ayes, a nd K-Nearest Neighbors (KNN)—were trained and evaluated using accuracy, precision, recall, and F1-score. LSTM achieved the best balance between diabetic classes, with an F1-score of 0.96 and overall precision of 97%. The system delivers scalable clinical assistance by creating eating plans using user-provided metrics (BMI and glucose) together with engineered diet components; these parameters are incorporated at inference to enable real-time personalization. Configuration f ollows recommendations suitable for both diabetic types and individual patients. This research demonstrates artificial intelligence’s potential to improve chronic disease management and lays foundations for future wearable device integration with real-time health monitoring systems. Future work will include continuous glucose monitor integration, mobile deployment, clinical validation studies, and expansion to culturally diverse meal corpora to assess long-term adherence and outcomes.