Skip to content
SDGs Center UDINUS

Evidence record · UDINUS · SDGs

Enhancing Nutritional Status Prediction through Attention-Based Deep Learning and Explainable AI

Rektorat UDINUS 2025 Published 2026-10-30

Evidence record · UDINUS · SDGs

Unit
Rektorat UDINUS
Period
2025
Goals
Record
48c75c89-88ed-4353-9b0f-2cd1d1ae18d4
Published
2026-10-30

Universitas Dian Nuswantoro conducted a study addressing accurate and interpretable prediction of child malnutrition. The activity involved researchers Santoso, Heru Agus; Dewi, Nur Setiawati; Haw, Su-Cheng; Pambudi, Arga Dwi; and Wulandari, Sari Ayu. The study identified the lack of transparency in existing artificial intelligence models as a challenge for clinical adoption. It introduced a deep learning framework enhanced with Multi-Head Attention, or MHA, for nutritional status prediction.

The activity included a direct head-to-head comparison of CNN-MHA and LSTM-MHA. The comparison evaluated spatial feature learning and sequential dependency modeling in structured anthropometric tabular data. The framework integrated advanced preprocessing techniques, feature selection, and Explainable AI using SHAP. These elements supported clinically aligned and transparent predictions. Experimental results were obtained from a dataset containing 9605 samples. CNN-MHA achieved 99.08% accuracy, while LSTM-MHA achieved 98.91% accuracy. The results confirmed that spatial modeling was better suited to this dataset type.

SHAP-based feature attribution identified WHO-standard z-scores as the most influential predictors. This result enhanced the credibility of the model for clinical application. The study also introduced an IoT-enabled anthropometric data acquisition system to enhance real-time monitoring and scalability. Its methodological focus covered nutritional status prediction, feature prioritization, accuracy, and interpretability. The activity addressed the relationship between high-accuracy artificial intelligence and clinical transparency and presented a scalable, explainable framework for public health interventions. The study stated that future research should explore multi-modal data integration to improve generalizability and real-world applicability. Through this activity, UDINUS researchers contributed a framework for explainable nutritional monitoring focused on child malnutrition and nutritional status prediction.

Reference https://www.scopus.com/pages/publications/105004674057?origin=resultslist