ESTIMATING THE HEALTH QUALITY OF CAMPUS MEALS USING ARTIFICIAL INTELLIGENCE

Campus dining plays a significant role in influencing student health, academic performance, and overall well-being. Many campus meal options do not provide easily accessible nutritional information at the point of purchase, which makes it challenging for students to make informed food choices. This study investigates the design and evaluation of three artificial intelligence (AI) approaches aimed at classifying and offering nutritional insights for campus meals. These approaches include: (1) Object detection using Roboflow/YOLOv8, (2) Image classification using MobileNetV2, and (3) optical character recognition (OCR)-based menu text analysis using Tesseract. A custom-built dataset was created using publicly available nutritional data from Nutritionix and images sourced from campus restaurants and academic collaborators. Each AI method was evaluated based on its ability to accurately identify food items and classify them as healthy, moderately healthy, or unhealthy based on a holistic set of nutritional metrics. The object detection model achieved the highest mean Average Precision (mAP@50) of 79%, while the OCR-based method reached a dish-level detection Accuracy of 85.94%. This research demonstrates the feasibility of AI-powered systems to facilitate real-time nutritional assessments and meal classifications. It contributes to the development of accessible health technology for university students, promoting better dietary choices and improving health outcomes.