Gender disparities and machine learning-based predictive modeling of school attendance disruption in somaliland: evidence from a national accessibility survey

School exclusion remains a substantial barrier to achieving universal and equitable education, particularly in fragile regions such as Somaliland. This study investigates the various determinants of school attendance disruptions, defined as children never having attended school, utilizing data from the 2022 Somaliland Education Accessibility Survey (N = 12,146 children aged 5–18). By employing a mixed-methods analytical approach, we integrated logistic regression with a range of machine learning classifiers (Decision Tree, Random Forest, Ridge Logistic Regression, and XGBoost) to identify key individual, household, and contextual factors influencing non-attendance and assess their predictive strength. The logistic regression analysis indicated that male children were significantly less likely to attend school compared to females (OR = 0.596, p < .001). Children in nomadic areas (reference), those from male-headed households (OR = 1.269, p < .001), and those from households reliant on borrowing for income were at a heightened risk of non-attendance. Conversely, urban (OR = 0.532, p < .001) and rural (OR = 0.721, p < .001) settings, along with more stable income sources such as remittances (OR = 0.292, p < .001), were associated with lower non-attendance rates. Notable regional differences were also observed. Although direct reports of financial and distance barriers were minimal, the XGBoost model, which demonstrated superior predictive performance (F1-score = 0.68 for out-of-school children), identified “inability to afford school” and “school distance barrier” as the most critical factors, along with urban location, age, and specific income sources. These findings underscore the complex interplay of gender, socio-economic status, geographic context, and underlying accessibility challenges in shaping school exclusion patterns in Somaliland. The study underscores the value of combining traditional statistical models for interpretability with advanced machine learning for enhanced predictive accuracy, providing essential insights for targeted policy interventions to improve educational access and equity.