Development of Tracking Features and Recommendation for Mother's Nutrition in the Genting Mobile Application to Prevent Stunting Using Vision Transformer (Case Study: PKK Bandung City)

Stunting is a chronic nutritional problem that has a serious impact on child growth and development in Indonesia. It not only affects physical stature but also impairs cognitive development and increases the risk of long-term socio-economic issues. Although several applications have been developed for stunting prevention, many do not provide efficient nutrition tracking features and personalized nutrition recommendations, which are crucial for early prevention of stunting. This research introduces the development of an Android mobile application called “Genting” that leverages the Vision Transformer (ViT) machine learning model for food image classification and Google Generative AI for providing tailored nutrition recommendations. The ViT model achieved an accuracy of 82.12% on the validation dataset after 20 epochs, with a loss value of 0.6725. To ensure model robustness, K-fold cross-validation with 5 folds was employed, resulting in an average accuracy of 95.93 Further testing demonstrated the model's ability to recognize various types of food with the highest probability reaching 99.99%, reflecting high confidence in its predictions. This high accuracy ensures more reliable food classification and improving efficiency in nutrition tracking. Although not publicly released, the Genting application underwent thorough testing in a restricted environment with Posyandu cadres to verify its functionality. This application is designed to provide innovative solutions in nutrition monitoring through advanced machine learning technology, offering more personalized and efficient nutrition recommendations. By doing so, it aims to support early stunting prevention efforts, addressing a significant public health challenge in Indonesia by improving the precision and effectiveness of nutritional interventions at a critical stage in child development.