Data-driven machine learning models for guiding the preparation of Mn-modified biochar and predicting Cd adsorption

Cadmium (Cd), a typical highly toxic heavy metal, poses serious risks to environmental safety and public health once it enters water bodies. Manganese-modified biochar (Mn-BC) can effectively remove Cd from wastewater. However, the need for complex batch experiments represents a major constraint on its development. The emergence of artificial intelligence provides a new approach to address this problem. In this study, six machine learning (ML) models were developed to predict the Cd adsorption performance of Mn-BC, aiming to reduce the reliance on complex experimental procedures. Among the six models, the extra trees regressor (ETR) model achieved high prediction accuracy (test R² = 0.970; root mean square error (RMSE) = 12.982). The ETR model provided insights into the preparation of Mn-BC: the results of feature importance analysis revealed that the initial Cd concentration, biochar dosage, and manganese content were the main contributors to the model output. Additionally, the one-way partial dependence plots revealed that the optimal heating time in the biochar preparation process was 2 h, and the optimal manganese loading reached approximately 10%. To evaluate the practical applicability of the model, Mn-BC was prepared for conducting reverse validation of the ETR model, and the error between the predicted and experimental values was less than 17%. The results of characterization analyses, including X-ray photoelectron spectroscopy (XPS) and X-ray diffraction (XRD), revealed that Mn-BC adsorbed Cd mainly via coordination exchange and surface precipitation. This research not only presents an effective model for predicting adsorbent performance but also provides data-supported insights into the associations between complex preparation and adsorption conditions, thereby offering effective guidance and new insights for practical adsorbent fabrication.