Scholarlink

<p>The increasing complexity of wastewater pollution demands robust predictive tools for optimizing treatment efficiency. Existing studies have largely focused on pollutant-specific systems, with limited efforts toward unified models applicable to diverse pollutants. This study addresses this gap by developing and evaluating machine learning (ML) models to predict pollutant removal efficiency (%) in biochar-based adsorption systems. A comprehensive dataset of 592 experimental records was compiled, encompassing parameters such as pH, pollutant concentration, biochar dose, contact time, and pyrolysis temperature. Four ensemble-based regressors—XGBRegressor, CatBoost, LightGBM, and HGBRegressor were optimized using the Yeo–Johnson transformation and GridSearchCV with 5-fold cross-validation. XGBRegressor achieved the highest training accuracy (R<sup>2</sup> = 0.867, RMSE = 8.738), while CatBoost demonstrated the best generalization on the test set (R<sup>2</sup> = 0.650, RMSE = 13.322), and is therefore selected as the preferred deployment model. Feature importance analysis indicated that pollutant type and biochar feedstock were the dominant predictors of removal efficiency. The developed framework provides a reliable tool for predicting adsorption behaviour across diverse pollutant classes and offers insights for optimizing operational conditions.</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

Ensemble Machine Learning Prediction of Multi-Pollutant Removal Efficiency in Biochar-Based Adsorption Systems

  • 期刊论文

  • 发表日期：

    2025年12月17日

查看全文
  • Gokulan Ravindiran,
  • K. Karthick,
  • Kolli Ramujee,
  • Mary Subaja Christo,
  • Madaminov Bekzod Allayarovich,
  • Hayitov Abdulla Nurmatovich,
  • Subhi A. Ali,
  • Gasim Hayder

摘要

The increasing complexity of wastewater pollution demands robust predictive tools for optimizing treatment efficiency. Existing studies have largely focused on pollutant-specific systems, with limited efforts toward unified models applicable to diverse pollutants. This study addresses this gap by developing and evaluating machine learning (ML) models to predict pollutant removal efficiency (%) in biochar-based adsorption systems. A comprehensive dataset of 592 experimental records was compiled, encompassing parameters such as pH, pollutant concentration, biochar dose, contact time, and pyrolysis temperature. Four ensemble-based regressors—XGBRegressor, CatBoost, LightGBM, and HGBRegressor were optimized using the Yeo–Johnson transformation and GridSearchCV with 5-fold cross-validation. XGBRegressor achieved the highest training accuracy (R2 = 0.867, RMSE = 8.738), while CatBoost demonstrated the best generalization on the test set (R2 = 0.650, RMSE = 13.322), and is therefore selected as the preferred deployment model. Feature importance analysis indicated that pollutant type and biochar feedstock were the dominant predictors of removal efficiency. The developed framework provides a reliable tool for predicting adsorption behaviour across diverse pollutant classes and offers insights for optimizing operational conditions.