Feature engineering for improved machine-learning-aided studying heavy metal adsorption on biochar.

Area of Science:

• Environmental Chemistry
• Materials Science
• Computational Chemistry

Background:

• Biochar is widely studied for heavy metal adsorption in wastewater.
• Machine learning (ML) models are increasingly used to predict biochar adsorption capacity.
• Existing ML studies prioritize algorithm development over feature engineering.

Purpose of the Study:

• To engineer biochar features for enhanced ML model performance.
• To improve the interpretability and generalizability of ML models for heavy metal adsorption.
• To identify key biochar features for predicting adsorption capacity.

Main Methods:

• Engineered elemental composition features of biochar on a mole basis.
• Developed a Gradient Boosting Regression (GBR) model.
• Introduced a new elemental ratio feature, (H-O-2N)/C, for model interpretation.
• Expanded model generalizability by incorporating external data.

Main Results:

• Achieved a test R² of 0.997 for the GBR model with engineered features.
• The (H-O-2N)/C ratio and biochar pH were identified as crucial predictors, replacing traditional metrics like cation exchange capacity (CEC).
• Validated model generalizability with external datasets, showing R² of 0.78 (without CEC/SSA) and 0.72 (experimental data).

Conclusions:

• Feature engineering significantly enhances ML model predictive performance and interpretability for biochar-based heavy metal adsorption.
• The proposed (H-O-2N)/C feature offers valuable insights into adsorption mechanisms.
• The developed ML model demonstrates strong generalizability and potential for practical application in wastewater treatment.