Prediction of H₂S adsorption capacity of biochar using rigorous machine learning frameworks

Area of Science:

• Environmental Science
• Materials Science
• Data Science

Background:

• Biochar is a sustainable adsorbent for hydrogen sulfide (H2S) removal.
• H2S adsorption capacity is influenced by complex interactions of biochar properties and operating conditions.
• Accurate prediction of H2S adsorption is crucial for optimizing biochar applications in gas purification.

Purpose of the Study:

• To develop and compare various machine learning (ML) models for predicting H2S adsorption capacity of biochar.
• To identify key factors influencing biochar's H2S adsorption performance.
• To provide a computationally efficient alternative to experimental methods for biochar screening.

Main Methods:

• Trained 277 data points from literature using Decision Tree, Random Forest, AdaBoost, KNN, CNN, SVR, and Ensemble Learning models.
• Utilized comprehensive inputs including biochar physicochemical properties, pyrolysis, and reaction conditions.
• Ensured model robustness via 5-fold cross-validation and outlier assessment; interpreted features using SHAP analysis.

Main Results:

• K-Nearest Neighbors (KNN) model achieved the highest accuracy (R^2 ≈ 0.94).
• Breakthrough time, specific surface area, gas humidity, and O/N ratio were identified as dominant factors.
• SHAP analysis provided interpretable insights into feature contributions.

Conclusions:

• ML models, particularly KNN, offer accurate and interpretable predictions of biochar's H2S adsorption capacity.
• This approach facilitates rapid screening and optimization of biochar for gas purification.
• The study highlights the potential of data-driven methods in advancing sustainable adsorbent technology.