Modeling and Forecasting Neonatal Mortality in Ethiopia: A Comparative Study Using Statistical, Machine Learning, and Deep Learning Approaches

Abstract Introduction: Ethiopia faces alarming stagnation in neonatal mortality rates (NMR) at ~27 deaths/1,000 live births, jeopardizing Sustainable Development Goal (SDG) 3.2 targets. Predictive analytics using advanced computational approaches remains underutilized for guiding interventions in low-resource settings. Methods: This comparative study employed national-level NMR data (1977–2023) from the WHO Global Health Observatory to forecast trends through 2034. Five models were evaluated: Auto Regressive Integrated Moving Average (ARIMA), Prophet, Random Forest, XGBoost, and Long Short-Term Memory (LSTM) networks. Data underwent logarithmic transformation and differencing to achieve stationarity. Model performance was assessed using Root Mean Squared Error (RMSE), Mean Absolute Percentage Error (MAPE), and R² under walk-forward validation. Results: The LSTM model demonstrated superior accuracy (RMSE: 0.0009; MAPE: 2.93%), outperforming statistical and machine learning approaches. Forecasts indicate marginal NMR decline to 27.71 (2030) and 27.13 (2034) deaths/1,000 live births, far exceeding the SDG 3.2 target of ≤12. Structural barriers include rural healthcare access gaps (80% population coverage), substandard perinatal care (20% facility compliance), and diagnostic delays contributing to 50% of deaths. Conclusion: Ethiopia requires a tenfold acceleration in NMR reduction to meet 2030 targets. Integrating LSTM-based forecasting into national health systems could enable proactive resource allocation. Urgent scale-up of Kangaroo Mother Care, emergency transport financing, and perinatal quality improvement is recommended. This study establishes LSTM as a transformative tool for neonatal mortality prediction in resource-limited contexts.