Scholarlink

This study aims to develop and predict a novel biological Soil Health Index (SHI) by integrating key biological variables using machine learning techniques. These variables capture essential microbial and enzymatic activities, providing a synthetic and integrative measure of soil health. The proposed SHI is intended to serve as a reliable proxy for agroecological models, decision-support systems, and environmental monitoring programs. We evaluated four supervised learning algorithms—Random Forest, XGBoost, CatBoost, and LightGBM—using the LUCAS_2018_soil_functions dataset. Model performance was assessed through three evaluation metrics: Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and the coefficient of determination ( \(R^2\) ). Among the tested models, CatBoost achieved the best predictive performance (MAE \(= 24.56\) , RMSE \(= 93.89\) , \(R^2 = 0.9046\) ), closely followed by Random Forest ( \(R^2 = 0.9028\) ). XGBoost ( \(R^2 = 0.8711\) ) and LightGBM ( \(R^2 = 0.8479\) ) showed slightly lower accuracy. The results indicate that gradient boosting methods, particularly CatBoost, have strong potential for accurately predicting soil biological health. This predictive capability can significantly contribute to sustainable agricultural management by enabling proactive monitoring and informed decision-making. The proposed approach offers a scalable and data-driven tool for enhancing soil conservation strategies and supporting long-term ecosystem resilience.

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

Soil Health Index Prediction Using Machine Learning: Comparison of Random Forest, XGBoost, CatBoost and LightGBM

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Pape ElHadji Abdoulaye Gueye,
  • Cherif Bachir Deme,
  • Adrien Basse

摘要

This study aims to develop and predict a novel biological Soil Health Index (SHI) by integrating key biological variables using machine learning techniques. These variables capture essential microbial and enzymatic activities, providing a synthetic and integrative measure of soil health. The proposed SHI is intended to serve as a reliable proxy for agroecological models, decision-support systems, and environmental monitoring programs. We evaluated four supervised learning algorithms—Random Forest, XGBoost, CatBoost, and LightGBM—using the LUCAS_2018_soil_functions dataset. Model performance was assessed through three evaluation metrics: Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and the coefficient of determination ( \(R^2\) ). Among the tested models, CatBoost achieved the best predictive performance (MAE \(= 24.56\) , RMSE \(= 93.89\) , \(R^2 = 0.9046\) ), closely followed by Random Forest ( \(R^2 = 0.9028\) ). XGBoost ( \(R^2 = 0.8711\) ) and LightGBM ( \(R^2 = 0.8479\) ) showed slightly lower accuracy. The results indicate that gradient boosting methods, particularly CatBoost, have strong potential for accurately predicting soil biological health. This predictive capability can significantly contribute to sustainable agricultural management by enabling proactive monitoring and informed decision-making. The proposed approach offers a scalable and data-driven tool for enhancing soil conservation strategies and supporting long-term ecosystem resilience.