David Satria Alamsyah, Aji Prasetya Wibawa, Adil Zakaria, Ahmad Ammar Musyaffa, Aldy Rahmat Yulianto, Agung Bella Putra Utama
Food allergy management remains challenging due to limited allergen disclosure and the complexity of ingredient compositions. Particularly in traditional cuisines such as Indonesian food, which constrains the effectiveness of existing digital decision-support approaches. This study aims to develop a health informatics–oriented Artificial Neural Network (ANN) framework for allergen classification and elimination diet recommendation. A large-scale Indonesian recipe dataset was constructed and annotated using 14 standardized allergen categories, and class imbalance was addressed through a hybrid strategy combining semi-manual rule-based data augmentation and Random Oversampling (ROS). The ANN model was evaluated using binary, ordinal, and multi-class classification scenarios with 5-fold cross-validation, supported by Random Search and Bayesian Optimization, and further examined using SHAP-based interpretability analysis. The results show that the proposed framework achieves consistently high performance across scenarios, including perfect performance in binary classification and substantial improvements in macro-averaged F1-scores in complex multi-class settings when data balancing strategies are applied. The findings also indicate that data-centric enhancement has a greater impact on performance robustness than hyperparameter tuning alone. Overall, the proposed ANN framework demonstrates potential as a reliable and interpretable component of health informatics–based elimination diet recommendation systems. © 2026 Elsevier Inc. All rights are reserved, including those for text and data mining, AI training, and similar technologies.
Department of Electrical Engineering and Informatics, Faculty of Engineering, Universitas Negeri Malang, Malang, Indonesia
