Unveiling sulfonamide adsorption on biochar using explainable machine learning.

Sulfonamide antibiotics are emerging aquatic contaminants due to environmental persistence and antimicrobial resistance risks. Biochar is a promising adsorbent for sulfonamide removal, but experimental optimization remains resource intensive. This study applies machine learning to predict biochar adsorption capacity using molecular properties of sulfonamides, adsorbent structural descriptors, and operating conditions. Artificial neural networks, Gaussian process regression, and gradient boosting were developed using grid-search cross-validation. Gradient boosting achieved the best performance (R2 = 0.997, RMSE = 6.06 mg g-1, MAE = 2.18 mg g-1). Explainability analysis using SHAP and partial dependence identified BET surface area as the dominant driver, with initial concentration and dosage as key operational factors, while pore diameter and pore volume showed secondary influence. This framework enables interpretable analysis of structure-performance relationships and provides guidance for biochar evaluation in water treatment applications.