Scholarlink

<p>Biochar application is a highly promising technology for the remediation of soils contaminated with heavy metals (HMs). However, the experimental optimization of biochar’s immobilization efficiency is often limited by significant time and cost requirements due to the complex interactions among various factors, such as soil properties, biochar properties, and experimental conditions. To address this challenge, this study proposes a hybrid machine learning (ML) framework integrating the Dung Beetle Optimizer (DBO), known for its strong global search capabilities, with eXtreme Gradient Boosting (XGBoost) to predict HM immobilization efficiency in biochar-amended soils. The DBO algorithm was employed to automatically tune the hyperparameters of three ML models (Random Forest (RF), Least Squares Boosting (LSBoost), and XGBoost). The results indicates that the optimized DBO-XGBoost model demonstrated superior predictive performance, achieving high accuracy (R<sup>2</sup> = 0.935) in estimating immobilization efficiency for four HMs (Hg, Cd, Pb, and Zn). Interpretability analyses using Shapley Additive Explanations (SHAP) and Partial Dependence Plots (PDPs) identified biochar carbon content and modifier content as the most influential features for immobilization efficiency. Optimal immobilization was observed at carbon contents of 30–40% (in weight) and modifier (e.g., sulfur, magnesium ferrite and MgO) contents exceeding 2%. The data-driven methodology proposed in this study provides practical insights for the design and optimization of biochar remediation strategies.</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

Predicting Heavy Metal Immobilization in Biochar-Amended Soils: A Hybrid Dung Beetle Optimizer-XGBoost Machine Learning Approach with Feature Analysis

  • 期刊论文

  • 发表日期：

    2026年6月18日

查看全文
  • MyongHyok Yang,
  • Erping Bi,
  • TaeChol Jang,
  • Chol Sok

摘要

Biochar application is a highly promising technology for the remediation of soils contaminated with heavy metals (HMs). However, the experimental optimization of biochar’s immobilization efficiency is often limited by significant time and cost requirements due to the complex interactions among various factors, such as soil properties, biochar properties, and experimental conditions. To address this challenge, this study proposes a hybrid machine learning (ML) framework integrating the Dung Beetle Optimizer (DBO), known for its strong global search capabilities, with eXtreme Gradient Boosting (XGBoost) to predict HM immobilization efficiency in biochar-amended soils. The DBO algorithm was employed to automatically tune the hyperparameters of three ML models (Random Forest (RF), Least Squares Boosting (LSBoost), and XGBoost). The results indicates that the optimized DBO-XGBoost model demonstrated superior predictive performance, achieving high accuracy (R2 = 0.935) in estimating immobilization efficiency for four HMs (Hg, Cd, Pb, and Zn). Interpretability analyses using Shapley Additive Explanations (SHAP) and Partial Dependence Plots (PDPs) identified biochar carbon content and modifier content as the most influential features for immobilization efficiency. Optimal immobilization was observed at carbon contents of 30–40% (in weight) and modifier (e.g., sulfur, magnesium ferrite and MgO) contents exceeding 2%. The data-driven methodology proposed in this study provides practical insights for the design and optimization of biochar remediation strategies.