Beyond health: A machine learning analysis of structural barriers to school attainment in Somalia

Abstract

In fragile states like Somalia, the link between poor health and educational exclusion is critical yet poorly understood. This study uses a novel machine learning approach to identify and rank the most significant barriers to school attendance. We analyzed nationally representative data from 10,511 children aged 6–18 in the 2022 Somalia Integrated Household Budget Survey (SIHBS). Ten supervised machine learning models were employed to predict school attendance, with the Random Forest model emerging as the top performer (AUC = 0.86). Contrary to conventional wisdom, our findings reveal a clear hierarchy of barriers where structural and demographic factors are the most powerful predictors. A child’s age (non-attendance rises from 6% in 6–10 year-olds to 25% in 15–18 year-olds), geographic region (non-attendance reaches 30.5% in Middle Shabelle), and residence type (nomadic children face triple the non-attendance risk of their peers) were the dominant determinants. These factors significantly outweighed the direct predictive power of individual health status or household poverty. The results indicate that while health is important, educational exclusion in Somalia is more fundamentally driven by where a child lives, how old they are, and their family’s mode of life. Policy interventions must therefore shift from broad, single-sector approaches to geographically-targeted, age-specific strategies. To effectively address the root causes of educational inequality, efforts should prioritize service delivery in marginalized regions and provide tailored support for nomadic and adolescent populations.