Scholarlink

<p>This study evaluates the comparative performance of machine-learning models in forecasting Somalia’s GDP growth using macroeconomic and environmental indicators over the period 1991–2022. The analysis applies Artificial Neural Networks (ANN), Convolutional Neural Networks (CNN), Support Vector Machines, Random Forest (RF), and Extreme Gradient Boosting (XGBoost) to assess their predictive capabilities in a data-constrained environment. Given the relatively small sample size 32 annual observations, particular attention is paid to model design and implementation to mitigate overfitting and ensure reliable predictive performance. Simplified model architectures and controlled hyperparameters are employed across all models, while validation is conducted using cross-validation to improve generalization. The results indicate that ANN and CNN outperform other techniques, achieving lower prediction errors and higher R<sup>2</sup> values, suggesting stronger predictive performance. In contrast, ensemble learning models such as Random Forest and XGBoost show comparatively lower performance, which may be attributed to the limited sample size and data constraints.The Support Vector Machine is applied in its regression form (Support Vector Regression, SVR), which is appropriate for modeling continuous variables such as GDP growth, ensuring conceptual consistency in the forecasting framework. Overall, the findings highlight the potential of machine-learning techniques for economic forecasting in fragile and data-limited contexts.</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

Comparative performance of machine learning models in forecasting Somalia’s GDP growth using macroeconomic and environmental indicators

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年6月29日

查看全文
  • Abdullahi Dahir Ibrahim,
  • Amir Mohamud Mohamed,
  • Bashir Mohamed Abdulle

摘要

This study evaluates the comparative performance of machine-learning models in forecasting Somalia’s GDP growth using macroeconomic and environmental indicators over the period 1991–2022. The analysis applies Artificial Neural Networks (ANN), Convolutional Neural Networks (CNN), Support Vector Machines, Random Forest (RF), and Extreme Gradient Boosting (XGBoost) to assess their predictive capabilities in a data-constrained environment. Given the relatively small sample size 32 annual observations, particular attention is paid to model design and implementation to mitigate overfitting and ensure reliable predictive performance. Simplified model architectures and controlled hyperparameters are employed across all models, while validation is conducted using cross-validation to improve generalization. The results indicate that ANN and CNN outperform other techniques, achieving lower prediction errors and higher R2 values, suggesting stronger predictive performance. In contrast, ensemble learning models such as Random Forest and XGBoost show comparatively lower performance, which may be attributed to the limited sample size and data constraints.The Support Vector Machine is applied in its regression form (Support Vector Regression, SVR), which is appropriate for modeling continuous variables such as GDP growth, ensuring conceptual consistency in the forecasting framework. Overall, the findings highlight the potential of machine-learning techniques for economic forecasting in fragile and data-limited contexts.