Hybrid time series modeling of primary school completers in Somaliland forecasts critical trends for sustainable educational planning
Description
Abstract Background: Accurate forecasting of student completion trends is an indispensable barometer of educational system efficacy and a critical driver for evidence-based policy planning. In post-conflict environments like Somaliland, operating at the intersection of rapid demographic expansion and acute resource scarcity, the Ministry of Education and Science faces a "secondary school bottleneck" where sudden surges in primary graduates threaten to overwhelm institutional capacity. Objective: This study aimed to evaluate the predictive accuracy of various classical univariate time-series models against sophisticated hybrid configurations to determine the most reliable framework for a decadal outlook. Methods: Utilizing a 29-year longitudinal dataset of annual primary school completers (1996–2024), the study compared 15 modeling configurations, including Autoregressive Integrated Moving Average (ARIMA), Exponential Smoothing (ETS), TBATS, Neural Network Autoregression (NNAR/ARNN), and Theta models. Data were chronologically partitioned into a training set (1996–2014) and a 10-year out-of-sample validation set (2015–2024). Model performance was benchmarked using Mean Absolute Percentage Error (MAPE), Root Mean Square Error (RMSE), and Theil’s U. Results: Diagnostic testing revealed that the series required second-order differencing (d=2), providing the first statistical proof that Somaliland’s primary completion rate is accelerating quadratically. While standalone ARIMA (MAPE = 4.63%) and ETS (MAPE = 4.65%) models performed well, the hybrid ARIMA–ETS–TBATS model achieved the highest precision with a test MAPE of 4.256% and a Theil’s U of 0.682. Conversely, complex neural networks (ARNN) yielded significantly higher error rates (MAPE = 26.63%), reinforcing the "small sample" constraint where parsimonious hybrid architectures outperform pure deep learning. Conclusion: Projections from the optimal hybrid model suggest that primary school completers will grow from 23,971 in 2025 to 33,435 by 2034. These results offer a validated roadmap for educational stakeholders to transition from reactive crisis management toward proactive "anticipatory governance," ensuring that systemic expansion is aligned with Sustainable Development Goal 4 (SDG 4) and Somaliland’s Education Sector Strategic Plan.
Files
Steps to reproduce
1. Software Requirements: Install R statistical software (version 4.4.0 or higher) and a standard IDE such as RStudio. 2. Required R Packages: Ensure the following specialized packages are installed in your R environment before running the script: forecast, tseries, TSstudio, forecastHybrid, and Metrics. 3. Environment Setup: Download the raw historical dataset (Somaliland primary school completers) and the provided R script into the same local working directory on your machine. 4. Execution: Open the R script in RStudio. Set your working directory to the folder containing the downloaded dataset (e.g., using the setwd() function or the RStudio interface). 5. Replication Workflow: Run the provided R script sequentially. The code is structurally commented and will automatically execute the following methodological phases: Phase A (Preprocessing): Ingestion of the raw data, time-series formatting, and chronological partitioning into the training set (1996–2014) and testing/validation set (2015–2024). Phase B (Diagnostics): Execution of Augmented Dickey-Fuller (ADF) and Phillips-Perron (PP) tests, verifying the requirement for second-order differencing (d=2). Phase C (Model Training): Estimation of the 6 standalone univariate models (ARIMA, ETS, TBATS, Theta, NNAR, ARFIMA) and the 9 equal-weight hybrid ensemble configurations. Phase D (Evaluation): Calculation of predictive accuracy metrics (RMSE, MAE, MAPE, sMAPE, and Theil’s U) on the out-of-sample validation set to replicate Tables 3 and 4. Phase E (Forecasting): Generation of the final 10-year decadal forecasts (2025–2034) utilizing the optimal ARIMA-ETS-TBATS hybrid model, and reproduction of the study's trajectory plots (Figures 1-4).
Institutions
- Amoud UniversityAwdal, Boorama