Neonatal Mortality Forecasting in Somalia: A Comparative Time-Series Analysis of Single and Hybrid Models

Abstract Background: Neonatal mortality remains a major public health challenge in Somalia, where fragile health systems, recurrent humanitarian crises, and limited access to quality maternal and newborn healthcare continue to contribute to preventable neonatal deaths. Reliable forecasting of neonatal mortality trends is essential for evidence-based policymaking, resource allocation, and monitoring progress toward Sustainable Development Goal (SDG) 3.2. Methods: This longitudinal time-series study analyzed annual neonatal mortality rate (NMR) data for Somalia from 1983 to 2023, obtained from the World Bank Open Data repository. Six individual forecasting models—Autoregressive Integrated Moving Average (ARIMA), Exponential Smoothing State Space (ETS), TBATS, Theta, Autoregressive Fractionally Integrated Moving Average (ARFIMA), and Neural Network Autoregression (NNAR)—together with ten hybrid model combinations were developed and compared. The dataset was divided into training (1983–2015) and testing (2016–2023) periods. Stationarity was evaluated using the Phillips–Perron and KPSS tests, with first-order differencing applied where appropriate. Model performance was assessed using Mean Absolute Percentage Error (MAPE), Symmetric Mean Absolute Percentage Error (sMAPE), and Theil’s U statistic. Results: Among the individual models, NNAR achieved the highest forecasting accuracy, producing the lowest prediction errors (MAPE = 4.84%, sMAPE = 0.0467, Theil’s U = 2.98). The Theta model ranked second, whereas ARIMA and ETS demonstrated moderate predictive performance, and ARFIMA showed the weakest results. Among the hybrid approaches, the Theta–NNAR model provided the best overall performance (MAPE = 6.11%). Forecasts for 2024–2030 suggest that neonatal mortality in Somalia will likely stabilize at approximately 35 deaths per 1,000 live births, although widening prediction intervals indicate increasing uncertainty over time. Conclusions: The findings demonstrate that nonlinear forecasting approaches, particularly NNAR, outperform conventional statistical models in capturing the complex dynamics of neonatal mortality in fragile settings. Although neonatal mortality is projected to remain relatively stable over the forecast period, the growing uncertainty highlights the importance of strengthening resilient, data-driven health planning and surveillance systems. These results provide valuable evidence to support targeted neonatal health interventions and accelerate progress toward achieving SDG 3.2 in Somalia.