Effects of physical properties on the heavy metal adsorption of biochar via a robust approach

• Developing a comprehensive machine–learning approach that relates biochar physicochemical attributes to metal–uptake efficiency. • SHAP interpretation revealed C 0 and CEC as dominant determinants, while surface pH exerted inverse influence on adsorption. • CNN exhibiting the highest overall predictive stability. Heavy‑metal contamination of soils and aqueous environments poses critical ecological and health risks, necessitating efficient sorbents for remediation. This study addresses the problem of unpredictable adsorption behavior of biochar by developing a comprehensive machine‑learning approach that relates its physicochemical attributes to metal‑uptake efficiency. A robust dataset of 380 experiments encompassing diverse biomass origins and preparation conditions was assembled to quantify this relationship using descriptors including elemental ratios, pH, cation‑exchange capacity (CEC), surface area, and structural charge. Eight algorithms (Decision Tree, AdaBoost, Random Forest, K‑Nearest Neighbor, Ensemble Learning, Convolutional Neural Network, Support Vector Regression, and Multilayer Perceptron) were evaluated through 5‑fold cross‑validation and optimized by hyperparameter tuning. Statistical indicators (R 2 , MSE, AARE%) and graphical diagnostics confirmed the CNN model as the most reliable predictor (R 2 = 0.991, MSE = 0.00148), capturing nonlinear physicochemical patterns with minimal overfitting. SHAP interpretation revealed C 0 and CEC as dominant determinants, while surface pH exerted inverse influence on adsorption. The hierarchical feature effects emphasize charge‑controlled and diffusion‑dependent mechanisms rather than morphological properties. The approach provides interpretable, transferable insight into how compositional and activation parameters govern heavy‑metal retention by biochars under varying conditions. Hence, the developed predictive framework not only advances modeling precision but also supports rational design of tailored biochars for environmental detoxification applications.