Scholarlink

<p>Arsenic contamination in water poses significant global health and environmental risks. While metal-modified biochar shows promise for arsenic removal, the rational design of high-performance adsorbents remains challenging due to complex parameter interactions. This study develops an interpretable machine learning (ML) framework to predict and analyze arsenic (As) adsorption on metal-modified biochar under batch-scale conditions. By integrating Lasso-based feature selection with heterogeneous ensemble learning, the constructed Lasso-MLP-RF model achieved high predictive accuracy (test set R² = 0.971, RMSE = 1.288&#xa0;mg/g) based on 686 experimental datasets. SHAP analysis indicates that the initial arsenic concentration (C<sub>0</sub>) and metal loadings (Fe and Mn) are among the most influential factors governing adsorption capacity, while partial dependence plots reveal nonlinear relationships and model-indicated interaction patterns among key parameters. This work establishes a data-driven foundation for understanding adsorption behavior and offers a useful reference for future material design optimization within the studied domain. The findings primarily yield theoretical-level insights, with further validation required for direct field applications.</p> Graphical 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

An interpretable machine learning framework for predicting and analyzing arsenic adsorption on metal-modified biochar

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年1月25日

查看全文
  • Ziwei Tang,
  • Bo Liang

摘要

Arsenic contamination in water poses significant global health and environmental risks. While metal-modified biochar shows promise for arsenic removal, the rational design of high-performance adsorbents remains challenging due to complex parameter interactions. This study develops an interpretable machine learning (ML) framework to predict and analyze arsenic (As) adsorption on metal-modified biochar under batch-scale conditions. By integrating Lasso-based feature selection with heterogeneous ensemble learning, the constructed Lasso-MLP-RF model achieved high predictive accuracy (test set R² = 0.971, RMSE = 1.288 mg/g) based on 686 experimental datasets. SHAP analysis indicates that the initial arsenic concentration (C0) and metal loadings (Fe and Mn) are among the most influential factors governing adsorption capacity, while partial dependence plots reveal nonlinear relationships and model-indicated interaction patterns among key parameters. This work establishes a data-driven foundation for understanding adsorption behavior and offers a useful reference for future material design optimization within the studied domain. The findings primarily yield theoretical-level insights, with further validation required for direct field applications.

Graphical Abstract