Forecasting unemployment in Somalia using machine learning models
- 1Department of Economics, Faculty of Economics and Management, Hormuud University
- 2Department of Economics, Faculty of Social Sciences, Banaras Hindu University (BHU)
Abstract
Unemployment remains a significant socioeconomic challenge in Somalia, underscoring the need for reliable forecasting tools to inform labor market planning and policy development. This study evaluates the performance of three machine learning models, such as Random Forest, XGBoost, Support Vector Regression (SVR), and a traditional ARIMA benchmark model, in forecasting Somalia’s unemployment rate using macroeconomic, investment, trade, and sectoral data over the period 1991–2021. Model performance is assessed using Mean Squared Error (MSE), Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), and the coefficient of determination (R²). The results show that the Random Forest model outperforms the other competing machine learning algorithms, as well as the benchmark method of ARIMA, in this study, indicating that it provides the most accurate unemployment forecasts throughout the period of study, achieving the lowest forecasting errors (MSE = 0.003, RMSE = 0.058, MAE = 0.050) and the highest explanatory power (R² = 0.92). SVR demonstrates moderate performance with MSE = 0.007, RMSE = 0.086, MAE = 0.067, and R² = 0.82, while XGBoost records the weakest results MSE = 0.009, RMSE = 0.094, MAE = 0.085, R² = 0.78. In contrast, the forecasted performance of the ARIMA model is the lowest amongst all models, having the following metrics for MSE, RMSE, MAE, and R² respectively: 0.119, 0.345, 0.332, and 0.31. This means that 31% of the change in unemployment is explained by ARIMA, and it generates considerably higher forecasting errors compared to machine learning algorithms. To enhance interpretability, SHapley Additive exPlanations (SHAP) are applied, revealing that labor force participation, foreign direct investment, and inflation are the most influential determinants of unemployment in Somalia, all exhibiting a negative relationship with unemployment. Other variables, such as domestic investment and imports, show moderate effects, while exports, agriculture, GDP growth, and industry have relatively limited direct influence. Overall, the findings confirm the superiority of Random Forest models for unemployment forecasting in fragile economic environments and provide policy-relevant insights for addressing labor market challenges in Somalia.
Related
- Flood susceptibility mapping using ensemble machine learning models and SHAP analysis in the Shebelle River Basin, Somalia
- Hybrid machine learning framework for drought forecasting: integration of LSBoost, RQNN, CNN-RNN, CNN-LSTM and advanced data decomposition
- Advancing Near-Term Water Quality Forecasting with Explainable Machine Learning and Probabilistic Uncertainty Quantification
- Station-specific variability and short-term forecasting of a particulate air-quality index using machine learning in Charleroi, Belgium
- Weather forecasting in a changing climate: the rise of AI and Machine learning?
BibTeX
@article{mohamed2026forecasting,
title = {{Forecasting unemployment in Somalia using machine learning models}},
author = {Amir Mohamud Mohamed and Abdullahi Dahir Ibrahim},
journal = {Discover Sustainability},
year = {2026},
month = jul,
doi = {10.1007/s43621-026-04175-9},
url = {https://doi.org/10.1007/s43621-026-04175-9},
note = {CC BY},
}
RIS
TY - JOUR
TI - Forecasting unemployment in Somalia using machine learning models
AU - Amir Mohamud Mohamed
AU - Abdullahi Dahir Ibrahim
PY - 2026
DA - 2026/07/01
JO - Discover Sustainability
DO - 10.1007/s43621-026-04175-9
UR - https://doi.org/10.1007/s43621-026-04175-9
AB - Unemployment remains a significant socioeconomic challenge in Somalia, underscoring the need for reliable forecasting tools to inform labor market planning and policy development. This study evaluates the performance of three machine learning models, such as Random Forest, XGBoost, Support Vector Regression (SVR), and a traditional ARIMA benchmark model, in forecasting Somalia’s unemployment rate using macroeconomic, investment, trade, and sectoral data over the period 1991–2021. Model performance is assessed using Mean Squared Error (MSE), Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), and the coefficient of determination (R²). The results show that the Random Forest model outperforms the other competing machine learning algorithms, as well as the benchmark method of ARIMA, in this study, indicating that it provides the most accurate unemployment forecasts throughout the period of study, achieving the lowest forecasting errors (MSE = 0.003, RMSE = 0.058, MAE = 0.050) and the highest explanatory power (R² = 0.92). SVR demonstrates moderate performance with MSE = 0.007, RMSE = 0.086, MAE = 0.067, and R² = 0.82, while XGBoost records the weakest results MSE = 0.009, RMSE = 0.094, MAE = 0.085, R² = 0.78. In contrast, the forecasted performance of the ARIMA model is the lowest amongst all models, having the following metrics for MSE, RMSE, MAE, and R² respectively: 0.119, 0.345, 0.332, and 0.31. This means that 31% of the change in unemployment is explained by ARIMA, and it generates considerably higher forecasting errors compared to machine learning algorithms. To enhance interpretability, SHapley Additive exPlanations (SHAP) are applied, revealing that labor force participation, foreign direct investment, and inflation are the most influential determinants of unemployment in Somalia, all exhibiting a negative relationship with unemployment. Other variables, such as domestic investment and imports, show moderate effects, while exports, agriculture, GDP growth, and industry have relatively limited direct influence. Overall, the findings confirm the superiority of Random Forest models for unemployment forecasting in fragile economic environments and provide policy-relevant insights for addressing labor market challenges in Somalia.
SN - 2662-9984
ER -