Integrated machine learning and response surface methodology for screening and optimization of high-performance Mg-modified bamboo biochar for phosphorus adsorption in water

Biochar adsorption is an effective method for removing excessive phosphorus from water bodies. In this study, the phosphorus adsorption capacity of biochar was investigated using a machine learning (ML) approach. A dataset consisting of 762 previously published data points from 122 types of biochar was analyzed, with the Random Forest model yielding the best performance (test R² = 0.98, test RMSE = 13.79). The predictions were validated through conducting laboratory experiments on 17 Mg-modified bamboo biochar samples (R² = 0.97, RMSE = 2.8, Q_max = 235.31 mg/g). The optimal preparation conditions were found to be a pyrolysis temperature of 700 °C, pyrolysis time of 3 h, and MgCl₂ concentration of 3 mol/L. X-ray photoelectron spectroscopy confirmed that adsorption performance is positively correlated with Mg content. Kinetic and isotherm tests (Langmuir Q_max = 240.55 mg/g) further supported these findings, providing insights for the preparation of high-performance Mg-modified bamboo biochar.