Scholarlink

<p>This study develops and comparatively evaluates time-series models to forecast the annual number of certified high school graduates in Somaliland. Using 25&#xa0;years of national administrative data (2000–2024), twelve forecasting models were estimated, including six univariate specifications (ARIMA, ETS, TBATS, Theta, NNAR, ARFIMA) and six hybrid combinations. Model performance was assessed using strict out-of-sample validation over a 6-year test period (2019–2024) and evaluated with MAE, RMSE, MAPE, and MASE metrics. Results indicate that the single TBATS model achieved the most consistent predictive accuracy (MASE = 1.13; MAPE = 5.36%), outperforming both traditional ARIMA-based and hybrid specifications. Increased model complexity did not improve forecast performance in this short annual series. The selected model projects continued growth in certified graduates, rising from approximately 16,366 in 2025 to over 25,000 by 2034, with widening prediction intervals reflecting forecast uncertainty. The findings suggest that parsimonious models may outperform hybrid complexity in data-constrained macro-educational contexts. The study contributes to educational forecasting research by extending time-series methodologies to system-level planning in a post-conflict setting.</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

Forecasting annual high school completers in Somaliland using univariate and hybrid time series models

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年5月29日

查看全文
  • Mustafe Khadar Abdi,
  • Jibril Abdikadir Ali,
  • Tawakale Abdi Ali,
  • Abdisalan Hassan Muse,
  • Mukhtaar Axmed Cumar

摘要

This study develops and comparatively evaluates time-series models to forecast the annual number of certified high school graduates in Somaliland. Using 25 years of national administrative data (2000–2024), twelve forecasting models were estimated, including six univariate specifications (ARIMA, ETS, TBATS, Theta, NNAR, ARFIMA) and six hybrid combinations. Model performance was assessed using strict out-of-sample validation over a 6-year test period (2019–2024) and evaluated with MAE, RMSE, MAPE, and MASE metrics. Results indicate that the single TBATS model achieved the most consistent predictive accuracy (MASE = 1.13; MAPE = 5.36%), outperforming both traditional ARIMA-based and hybrid specifications. Increased model complexity did not improve forecast performance in this short annual series. The selected model projects continued growth in certified graduates, rising from approximately 16,366 in 2025 to over 25,000 by 2034, with widening prediction intervals reflecting forecast uncertainty. The findings suggest that parsimonious models may outperform hybrid complexity in data-constrained macro-educational contexts. The study contributes to educational forecasting research by extending time-series methodologies to system-level planning in a post-conflict setting.