Enhancing lead adsorption capacity prediction in biochar: a comparative study of machine learning models and...

Area of Science:

• Environmental Science
• Materials Science
• Data Science

Background:

• Machine learning models for predicting lead adsorption in biochar are underdeveloped.
• Existing models lack sufficient accuracy for practical environmental applications.

Purpose of the Study:

• To develop and compare machine learning models (BPNN and RF) for predicting lead adsorption in biochar.
• To identify key biochar preparation features influencing lead adsorption.
• To enhance model accuracy through feature selection and optimization.

Main Methods:

• Developed back-propagation neural network (BPNN) and random forest (RF) models.
• Utilized features: preparation temperature (T), specific surface area (BET), relative carbon content (C), H/C, O/C, N/C ratios, and cation exchange capacity (CEC).
• Applied feature selection and particle swarm optimization to improve the RF model.

Main Results:

• The RF model demonstrated superior performance over BPNN, with a 10% R² improvement.
• Optimized RF model achieved an 8.3% R² increase, up to 56.1% RMSE reduction, and 55.7% MAE reduction.
• Feature importance ranking: CEC > C > BET > O/C > H/C > N/C > T, emphasizing CEC's role.

Conclusions:

• Machine learning, particularly RF with optimization, offers a powerful tool for predicting lead adsorption in biochar.
• Cation exchange capacity (CEC) is a critical feature for lead adsorption prediction.
• Findings support developing optimized biochar for effective lead removal in water remediation.