DOI: CU-CIARI-0000000126

Authors: Ismail Mahamoud Yousuf; Suhaib Mohamed Kahie Seiman; Ahmed Farah Daarood; Nathan Mongute Nyamweya

Description: Background: Neonatal mortality remains a major public health challenge in Somalia, where conflict, weak health systems, and limited maternal and newborn healthcare continue to affect survival. This study aimed to forecast neonatal mortality trends using single and hybrid time-series models to identify the most accurate approach for supporting evidence-based health planning. Methods: Annual neonatal mortality data for Somalia (1983–2023) were obtained from the World Bank Open Data repository. Six single models (ARIMA, ETS, TBATS, Theta, ARFIMA, and NNAR) and ten hybrid model combinations were developed and evaluated. Forecast accuracy was assessed using Mean Absolute Percentage Error (MAPE), Symmetric Mean Absolute Percentage Error (sMAPE), and Theil’s U statistic. Results: The Neural Network Autoregression (NNAR) model achieved the highest predictive accuracy among the single models, while the Theta–NNAR hybrid model was the best-performing hybrid approach. Forecasts for 2024–2030 suggest that neonatal mortality in Somalia is likely to stabilize at approximately 35 deaths per 1,000 live births, although prediction uncertainty increases over longer forecast periods. Conclusions: Nonlinear and hybrid forecasting models provide reliable tools for predicting neonatal mortality trends in fragile settings. The findings offer valuable evidence to support maternal and newborn health planning, resource allocation, and progress toward Sustainable Development Goal (SDG) 3.2 in Somalia.

Version: v1.0

Resource Type: Dataset

Language: English

Publication Date: 2026-05-17

Publisher: CIARI

License: Creative Commons Attribution 4.0

Keywords: Neonatal mortality, Time-series analysis, Forecasting, Neural Network Autoregression (NNAR), Hybrid models, Forecast accuracy, Somali

Journal: Open Research Africa

Volume/Issue: 9 / 32

Pages: 1-20

---

Files