Artificial and Convolutional Neural Network Architectures for Childhood Stunting Classification: Design, Evaluation, and Optimization

Childhood stunting is a global health challenge, affecting 148 million children under 5 in 2022. It is a key indicator of chronic malnutrition, often driven by inadequate nutrition, recurrent infections, and socio-economic disparities. In Egypt, 22% of children under 5 are stunted, leading to cognitive delays, poor educational outcomes, and long-term economic losses. Innovative and interdisciplinary approaches are essential to address this issue. This research aims to enhance the detection of growth abnormalities in children using advanced machine learning models, including Artificial Neural Networks (ANN), Convolutional Neural Networks (CNN), and Autoencoder-based architectures. The study utilized secondary data from Demographic Health Surveys (DHS) conducted in Egypt between 2005 and 2014, comprising 37,051 records. Key maternal and child characteristics were analyzed to calculate Height-for-Age Z -scores (HAZ) and Weight-for-Age Z -scores (WAZ). A 70:30 train-test split was applied, and dropout layers were used to prevent overfitting during model training. The ANN model achieved 99.5% accuracy, with a precision of 97.2% for normal cases and 95.4% for severely stunted cases. The CNN model achieved lower accuracy (68%) but provided valuable insights into spatial growth patterns. Autoencoder-enhanced models (e.g., AE + ANN) demonstrated moderate performance, with AE + ANN achieving 77.2% accuracy. Misclassification rates for stunted versus severely stunted cases reached up to 14%. This study demonstrates the potential of machine learning models in early detection and intervention for childhood stunting. By leveraging automated, data-driven approaches, healthcare providers can make evidence-based decisions, allocate resources effectively, and improve child health outcomes in Egypt and beyond.