Unraveling the role of Fe in As(III & V) removal by biochar via machine learning exploration

Biochar adsorption is a conspicuous technology for As remediation, and Fe modification into biochar is deemed an efficient approach to enhance As removal. Recently, immense research resources have been dedicated to the synthesis of applicable adsorbents. Herein, based on biochar characteristics and reaction conditions reported in literature, machine learning was applied to model As adsorption capacity onto pristine biochar and Fe-modified biochar to guide adsorbent design and process optimization. The random forest algorithm was employed, and four essential hyperparameters were tuned using an iterative method. With R2 of 0.9714, the optimized model presented high prediction accuracy for As adsorption capacity. Moreover, the feature importance analysis implied that the initial As concentration, H/C atomic ratio, Fe content, and surface area played dominant roles in this multifactorial process. Furthermore, the indirect impact of impregnated Fe on As removal was revealed as altering biochar properties (e.g., O/C atomic ratio, surface area). The information mining behind the model helped in understanding adsorption mechanisms and supported the rational design of engineered biochar to remove As and the enhancement of process operations without repetitive experiments. Also, this study provided a generic reference for the application of machine learning in both laboratory investigations and practical applications.