Prediction of H ₂ S adsorption capacity of biochar using rigorous machine learning frameworks

Biochar is a promising, sustainable adsorbent for hydrogen sulfide (H₂S) removal, yet its adsorption capacity is governed by complex interactions among material properties, preparation conditions, and operating parameters. In this study, we develop and systematically compare a suite of machine learning (ML) models including Decision Tree, Random Forest, AdaBoost, K-Nearest Neighbors (KNN), Convolutional Neural Network (CNN), Support Vector Regression (SVR), and an Ensemble Learning scheme to predict the H₂S adsorption capacity of biochar. The models are trained on 277 experimental data points collected from the literature, using a comprehensive set of inputs that includes physicochemical properties (specific surface area, mass percentages of C, O, and N, C/N, O/N, (O + N)/C, total pore volume, and average pore diameter), pyrolysis conditions (temperature and time), and reaction conditions (gas humidity, adsorption temperature, H₂S concentration, gas flow rate, and breakthrough time). Model robustness is ensured through 5-fold cross-validation and rigorous outlier assessment using the Leverage (Williams) method, while SHapley Additive exPlanations (SHAP) are applied to interpret feature contributions. Among all algorithms, KNN emerges as the best-performing model, achieving the highest coefficient of determination (<i>R</i>² ≈ 0.94) and the lowest mean squared error and average absolute relative error on the full dataset. Sensitivity and SHAP analyses consistently identify breakthrough time as the dominant factor controlling adsorption capacity, followed by specific surface area, gas humidity, and oxygen-to-nitrogen ratio. These findings demonstrate that combining diverse ML architectures with robust statistical validation provides an accurate, interpretable, and computationally efficient alternative to conventional experimental determination of H₂S adsorption capacity, facilitating rapid screening and optimization of biochar-based gas purification systems.