Using deep learning to monitor children’s nutritional status for child stunting identification

Abstract

Introduction Children's nutritional health remains a major public health concern, and accurate early identification of growth related risks is essential for timely screening and intervention. This study proposes a deep learning based framework for monitoring children's nutritional status for child stunting identification. Methods The task is formulated as a supervised binary classification problem, in which each child is represented by routinely available demographic and anthropometric variables, including age, sex, height, weight, and body mass index. Based on these structured inputs, the proposed Intelligent Nutritional Monitoring Model (INMM) learns discriminative latent representations and predicts whether a child is stunted. The model adopts a compact feature extraction architecture with feature fusion and hierarchical attention, enabling effective modeling of informative interactions among child level attributes while maintaining computational efficiency. To evaluate the proposed method, experiments are conducted on two public health survey datasets, NHANES and NFHS 5, under a unified preprocessing, training, and evaluation protocol. Results and Discussion Comparative results against traditional machine learning, mainstream deep learning, lightweight tabular modeling, and transformer based tabular baselines show that the proposed method achieves the strongest overall classification performance on both datasets while preserving a favorable effectiveness efficiency trade off. Additional ablation results further confirm the contribution of feature extraction, fusion, attention, and prediction design to the final performance. These findings indicate that the proposed framework provides an effective and scalable solution for data driven child stunting identification from structured health survey data.