Scholarlink

<p>Agriculture is vital for food security and economic development and optimizing crop selection based on data is key for sustainable farming. This study uses machine learning (ML) models to recommend suitable crops based on climate and soil data in Brazil, addressing a gap in the literature that often focuses on Kaggle data, localized data and individual soil components. Using data from the Brazilian Ministry of Agriculture and the National Institute of Meteorology (INMET), the study aggregates crop productivity data from 32 soil types and 95 crop types farmed in 3961 cities in Brazil with daily climate data. Various ML techniques were explored, including Gradient Boosting, Logistic Regression with Bagging, support vector machine (SVM) with Bagging, and Random Forest, all the ML methods tested different preprocessing methods. Synthetic Minority Over-sampling Technique (SMOTE) addressed class imbalance, and multithreading improved computational efficiency. Results showed that the Random Forest performed best with climate data alone (F1-score of 0.20), while adding soil types increased the score to 0.42. For city-specific data, Gradient Boosting achieved an F1-score of 0.70. Tailored models for specific cities outperformed generalized ones, highlighting the challenges of creating general models. Localized models demonstrated to be easier to train but have less generalizing power.</p> Graphic abstract <p></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

Machine learning for optimized crop selection: a data-driven approach considering climate and soil type data

  • 期刊论文

  • 发表日期：

    2026年1月8日

查看全文
  • Joao Henrique Gomes da Costa Cavalcanti,
  • Moisés Cirilo de Brito Souto,
  • Eduardo de Moura Oliveira Filho,
  • Erich Matos Rodrigues,
  • Yan Evangelista Barros,
  • Allan de Medeiros Martins

摘要

Agriculture is vital for food security and economic development and optimizing crop selection based on data is key for sustainable farming. This study uses machine learning (ML) models to recommend suitable crops based on climate and soil data in Brazil, addressing a gap in the literature that often focuses on Kaggle data, localized data and individual soil components. Using data from the Brazilian Ministry of Agriculture and the National Institute of Meteorology (INMET), the study aggregates crop productivity data from 32 soil types and 95 crop types farmed in 3961 cities in Brazil with daily climate data. Various ML techniques were explored, including Gradient Boosting, Logistic Regression with Bagging, support vector machine (SVM) with Bagging, and Random Forest, all the ML methods tested different preprocessing methods. Synthetic Minority Over-sampling Technique (SMOTE) addressed class imbalance, and multithreading improved computational efficiency. Results showed that the Random Forest performed best with climate data alone (F1-score of 0.20), while adding soil types increased the score to 0.42. For city-specific data, Gradient Boosting achieved an F1-score of 0.70. Tailored models for specific cities outperformed generalized ones, highlighting the challenges of creating general models. Localized models demonstrated to be easier to train but have less generalizing power.

Graphic abstract