Scholarlink

<p>Human fatalities in Somalia represent one of the world’s most protracted humanitarian crises, driven by multi-decadal conflict, political instability, and climate-induced shocks. While previous research primarily focuses on retrospective assessments, limited data-driven predictive frameworks hinder proactive humanitarian planning. This study fills this gap by evaluating the predictive performance of fifteen advanced time-series modeling frameworks to forecast fatality trajectories using a 30-year longitudinal dataset (1997–2026), noting that the terminal data point represents a projection based on available partial-year reports.&#xa0;Accounting for the constraints of a small annual sample size, the dataset was partitioned into training (1997–2016) and validation (2017–2026) phases to test six standalone models—including ARIMA, ETS, Theta, NNAR, ARFIMA, and TBATS—alongside nine hybrid ensemble configurations. Stationarity was validated via Augmented Dickey-Fuller (ADF) and Phillips-Perron (PP) tests, with performance benchmarked using RMSE, MAE, sMAPE, MASE, and Theil’s U metrics. Decadal analysis revealed a significant upward trajectory; total fatalities escalated from 3770 in the first decade (1997–2006) to 56,247 in the most recent decade (2017–2026), representing an alarming 1391% increase. While ARFIMA and Theta emerged as the superior standalone frameworks on specific error magnitudes (e.g., Theta sMAPE: 0.3653), the hybrid ARIMA-ETS-Theta (AET) model demonstrated the highest balanced predictive stability across all configurations (sMAPE: 0.3697; RMSE: 2551.36). The AET hybrid is adopted as the primary forecasting framework for this study to effectively balance autoregressive lags with trend-smoothing. Forecasts from this optimal hybrid model project a sustained high-intensity mortality cycle through 2036, with annual fatalities expected to fluctuate at historically elevated levels. These findings underscore the effectiveness of hybrid modeling in capturing the volatile dynamics of conflict data and provide a validated evidence base to align humanitarian interventions with Sustainable Development Goals 3 and 16. This research provides crucial, data-driven insights for Somali policymakers to transition from reactive crisis management toward proactive public health sustainability in a highly vulnerable nation.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Hybrid time series modeling of human fatalities in somalia forecasts critical trends for sustainable humanitarian planning

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月2日

查看全文
  • Mohamed Abdi Abdullahi,
  • Abdisalam Hassan Muse,
  • Yahye Hassan Muse,
  • Mukhtar Abdi Hassan,
  • Saralees Nadarajah

摘要

Human fatalities in Somalia represent one of the world’s most protracted humanitarian crises, driven by multi-decadal conflict, political instability, and climate-induced shocks. While previous research primarily focuses on retrospective assessments, limited data-driven predictive frameworks hinder proactive humanitarian planning. This study fills this gap by evaluating the predictive performance of fifteen advanced time-series modeling frameworks to forecast fatality trajectories using a 30-year longitudinal dataset (1997–2026), noting that the terminal data point represents a projection based on available partial-year reports. Accounting for the constraints of a small annual sample size, the dataset was partitioned into training (1997–2016) and validation (2017–2026) phases to test six standalone models—including ARIMA, ETS, Theta, NNAR, ARFIMA, and TBATS—alongside nine hybrid ensemble configurations. Stationarity was validated via Augmented Dickey-Fuller (ADF) and Phillips-Perron (PP) tests, with performance benchmarked using RMSE, MAE, sMAPE, MASE, and Theil’s U metrics. Decadal analysis revealed a significant upward trajectory; total fatalities escalated from 3770 in the first decade (1997–2006) to 56,247 in the most recent decade (2017–2026), representing an alarming 1391% increase. While ARFIMA and Theta emerged as the superior standalone frameworks on specific error magnitudes (e.g., Theta sMAPE: 0.3653), the hybrid ARIMA-ETS-Theta (AET) model demonstrated the highest balanced predictive stability across all configurations (sMAPE: 0.3697; RMSE: 2551.36). The AET hybrid is adopted as the primary forecasting framework for this study to effectively balance autoregressive lags with trend-smoothing. Forecasts from this optimal hybrid model project a sustained high-intensity mortality cycle through 2036, with annual fatalities expected to fluctuate at historically elevated levels. These findings underscore the effectiveness of hybrid modeling in capturing the volatile dynamics of conflict data and provide a validated evidence base to align humanitarian interventions with Sustainable Development Goals 3 and 16. This research provides crucial, data-driven insights for Somali policymakers to transition from reactive crisis management toward proactive public health sustainability in a highly vulnerable nation.