A Machine Learning-Based Mobile Application for Automatic Food Nutrition Identification

Abstract

Nutriverse is a mobile application designed to assist individuals in monitoring their daily nutritional intake through automated food recognition. The product specifically functions by identifying food items from images and estimating their nutritional composition in real time. The main issue addressed in this research is the difficulty users face in obtaining fast and accurate nutritional information, especially for Indonesian dishes that are not well represented in existing food-recognition datasets. This study aims to develop an accurate, user-friendly solution that enhances dietary monitoring while addressing the lack of localized datasets. The methodology incorporates Scrum for iterative development, the creation and annotation of an Indonesian food image dataset, and the deployment of a YOLOv12s model trained to detect food types and estimate nutrition. The results show reliable model performance with precision and recall scores above 0.7 and a mean Average Precision (mAP50–95) of approximately 0.6. In addition, system testing through User Acceptance Testing (UAT) ensures that all key functions operate stably and in accordance with user operational requirements. Usability evaluation using the System Usability Scale (SUS) resulted in a score of 76, placing the application in the "Good" category. Nutriverse provides a practical solution for accurate nutrition monitoring and contributes to technological innovation in the industry to support public health and well-being.