Explainable Machine Learning for Multiclass Classification of Concurrent Child Undernutrition in Ethiopia

Child undernutrition remains a major public-health challenge in Ethiopia, often occurring in concurrent forms that are clinically more severe than single deficits. We develop a supervised machine-learning framework to classify children into concurrent nutritional states using World Health Organization anthropometric indicators. Using baseline data from the Young Lives Cohort Study, we model seven observed nutritional categories under substantial class imbalance. Models were evaluated using imbalance-aware metrics, including Macro-F1, Balanced Accuracy, and ROC-AUC. Random Forest achieved the strongest overall performance and provided improved discrimination for concurrent undernutrition cate-gories. Explainability analysis using SHAP highlighted the importance of house-hold and caregiver-related factors. These findings demonstrate the potential of explainable machine-learning approaches for modeling concurrent undernutrition and provide a foundation for future longitudinal and multi-label extensions.