Phenolic pollutants are toxic and widespread, making their removal an urgent priority. Metal-modified biochar (MBC) shows great potential for non-radical activation of persulfate (PS), enabling efficient and low-carbon removal of phenolic pollutants. However, regulating the non-radical mechanism remains challenging due to the lack of systematic understanding of how metal-modification-related preparation parameters influence the reaction pathway. Herein, a machine learning (ML) framework focusing on metal modification parameters was introduced. Five ML algorithms were used to build binary classification models. XGBoost achieved the best performance, with an area under the receiver operating characteristic curve (AUC) of 0.711 on the independent test set and both precision and recall of 0.733 for identifying the non-radical-dominated mechanism. SHapley Additive exPlanations (SHAP) and partial dependence analysis reveal that high temperatures (>800 °C) and moderate heat treatment time (1.7–2.5 h) favor non-radical pathways, with Cu introduction and multi-metal synergy serving as key regulatory factors. Based on SHAP dependency analysis and metal combination statistics, two strategies for the targeted design of non-radical-dominated systems are proposed: (1) prepare Cu-modified biochar at 900 °C; (2) prepare Cu-Fe bimetallic-modified biochar at 600 °C. This work provides a data-driven theoretical framework and operational strategies for targeted regulation of the non-radical mechanism in the MBC/PS system, opening new avenues for the efficient and low-carbon treatment of phenol-containing wastewater.
Wei et al. (2026) studied this question.