Machine Learning-Based Prediction and Optimization of Heavy Metal Adsorption Performance of Biochar

Biochar has been extensively employed in wastewater treatment owing to its effectiveness in removing heavy metal ions. However, the relationships among biomass feedstock composition, pyrolysis conditions, and adsorption performance are highly nonlinear and difficult to quantify systematically across heterogeneous experimental studies. Random Forest (RF), Gradient Boosting Regression (GBR), and XGBoost (XGB) algorithms were employed to predict and optimize biochar yield, adsorption-related physicochemical properties, and adsorption capacity (qe). The models were developed using two literature-derived datasets comprising 431 samples for physicochemical property and yield prediction (Data 1) and 452 samples for adsorption capacity modeling (Data 2). Results indicated that XGB exhibits superior overall predictive performance across most tasks. Specifically, the single-target XGB models achieved coefficients of determination (R2) of 0.89–0.92 and root mean square errors (RMSE) of 0.02–0.07 on the test set, while the multi-target model attained an R2 of 0.83 with an RMSE of 0.05. Further analysis reveals that pyrolysis conditions exert a dominant influence on biochar yield, with pyrolysis temperature identified as the most critical factor. In contrast, the physicochemical properties of biochar and its adsorption performance are primarily governed by feedstock composition, particularly carbon and ash contents. Higher ash content impairs surface functionality and reduces adsorption capacity (qe), whereas increased carbon content and appropriately optimized pyrolysis conditions contribute to enhanced adsorption capacity. In addition to intrinsic biochar properties, the initial concentration of heavy metal ions in solution constitutes an important external factor influencing adsorption behavior. Feature importance analysis, SHAP analysis, and correlation analysis collectively elucidate the key factors affecting biochar characteristics and adsorption performance, as well as the interactions among these factors. These findings provide a data-driven basis for optimizing biochar production parameters and guiding experimental design for efficient Cu2+ and Pb2+ removal from wastewater.