Robust Data driven modeling of Cd(II) adsorption on biochar

• Modeling Cd(II) adsorption on biochar • carbon content, pyrolysis duration, and initial Cd(II) concentration exert dominant and directionally opposite effects on adsorption • CNN, AdaBoost, and RF achieved superior generalization Cadmium toxicity presents a critical challenge for water security, driving the search for efficient biochar adsorbents. This study develops predictive machine learning models for Cd(II) removal using a homogenized global dataset of 1,150 experimental records. Seven algorithms were optimized under five-fold cross-validation to estimate adsorption capacity from physicochemical and operational descriptors. Convolutional Neural Network (CNN), AdaBoost, and Random Forest (RF) achieved superior predictive consistency ( R 2 > 0.99 ), whereas SVR and MLP-ANN exhibited weak generalization. SHapley Additive exPlanations (SHAP) analysis revealed that carbon content and pyrolysis duration exert dominant negative effects on adsorption, contrasting with the positive influence of initial concentration. These opposing trends mechanistically highlight the trade-off between surface functionality and porosity. Overall, this work advances computational environmental research, offering a validated framework for the targeted design of biochars to maximize heavy metal sequestration.