<![CDATA[Background: Optimal health relies on balanced nutrition, but many struggle with food choices due to limited knowledge and restricted access to nutritionists. Existing solutions analyze food images or health texts separately, leaving a gap in integrating real-time image recognition with personalized medical inference.Aims: This study aims to develop a personalized food recommendation expert system combining Convolutional Neural Network (CNN)-based image classification with backward chaining reasoning, mapping food images and user health profiles (BMI, medical history, allergies) into structured IF–THEN rules.Methods: Using a software engineering R&D design, the system was built on a three-tier architecture guided by WHO and Indonesian Ministry of Health standards. Performance was evaluated via black-box testing, a 500-sample food image dataset, and synthetic user profile simulations (normal, obese, hypertensive, allergic).Result: The CNN model achieved 87% accuracy across 500 test images, exceeding the 85% target threshold. Simulations showed an 80% average alignment with official dietary standards, successfully automating caloric restrictions, low-sodium adjustments, and allergen eliminations.Conclusion: Combining computer vision with backward chaining inference effectively bridges the gap in automated personalized nutrition, serving as a self-service tool for the public and a decision support system for health professionals.]]>
Copyrights © 2026
