Scholarlink

<p>Accurate prediction of cereal production is critical for food security in regions facing climatic and resource challenges. This study develops a machine learning (ML) framework to forecast cereal production (measured in metric tons) in Algeria by integrating climatic (e.g., maximum temperature, number of hot days, rainfall) and agricultural (e.g., fertilizer consumption, equipped irrigation land, cereal land area) data from 1961 to 2022. The RRelieF method selected key predictors, and six input scenarios designed to isolate the contributions of climatic versus agricultural variables and their optimal combinations were evaluated Eight ML models, including a novel Gradient Boosting-Support Vector Machine (GB-SVM) hybrid, were assessed. Results showed that models integrating both data types outperformed others, with Gradient Boosting and the GB-SVM hybrid achieving the highest accuracy (testing R² up to 0.933, RMSE as low as 0.323). A SHAP analysis confirmed that equipped irrigation land, cereal land area, and temperature extremes were the most influential predictors. These findings underscore the value of integrating multidimensional data with advanced ML for robust agricultural forecasting, though performance may be constrained by historical data limitations. This framework provides policymakers and agricultural planners with an actionable tool for enhancing resource allocation and strategic planning.</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

Assessment of machine learning models for predicting cereal production in Algeria based on climatic and agricultural variables

  • 期刊论文

  • 发表日期：

    2025年11月10日

查看全文
  • Tarek Bouregaa

摘要

Accurate prediction of cereal production is critical for food security in regions facing climatic and resource challenges. This study develops a machine learning (ML) framework to forecast cereal production (measured in metric tons) in Algeria by integrating climatic (e.g., maximum temperature, number of hot days, rainfall) and agricultural (e.g., fertilizer consumption, equipped irrigation land, cereal land area) data from 1961 to 2022. The RRelieF method selected key predictors, and six input scenarios designed to isolate the contributions of climatic versus agricultural variables and their optimal combinations were evaluated Eight ML models, including a novel Gradient Boosting-Support Vector Machine (GB-SVM) hybrid, were assessed. Results showed that models integrating both data types outperformed others, with Gradient Boosting and the GB-SVM hybrid achieving the highest accuracy (testing R² up to 0.933, RMSE as low as 0.323). A SHAP analysis confirmed that equipped irrigation land, cereal land area, and temperature extremes were the most influential predictors. These findings underscore the value of integrating multidimensional data with advanced ML for robust agricultural forecasting, though performance may be constrained by historical data limitations. This framework provides policymakers and agricultural planners with an actionable tool for enhancing resource allocation and strategic planning.