Machine learning-based prediction of PAHs thermal desorption efficiency: Model optimization, boundary correction, and...

Area of Science:

• Environmental Engineering
• Machine Learning Applications
• Soil Remediation Technologies

Background:

• Predicting Polycyclic Aromatic Hydrocarbons (PAHs) removal efficiency during thermal desorption is complex due to interacting soil, contaminant, and operational factors.
• Existing methods struggle with nonlinear interactions, limiting the optimization and reliability of thermal desorption processes for contaminated soils.

Purpose of the Study:

• To develop and optimize machine learning (ML) models for accurate prediction of PAHs removal efficiency in thermal desorption.
• To enhance the physical plausibility and accuracy of predictions, especially for high removal efficiencies (>90%).
• To identify key parameters influencing PAHs thermal desorption and provide insights for process optimization.

Main Methods:

• Developed and optimized nine ML models, incorporating Bayesian hyperparameter tuning and boundary-aware target transformation (logit-based).
• Applied interpretability techniques (Shapley Additive Explanations, partial dependence) to identify dominant predictors.
• Validated model generalization using Monte Carlo simulations and stratified modeling based on pollutant species.

Main Results:

• Ensemble tree models, particularly CatBoost, achieved the highest prediction accuracy (R² = 0.9709) without overfitting.
• Reaction time, temperature, and boiling point-to-temperature ratio were identified as dominant predictors.
• Soil moisture and organic matter showed dual roles, supporting two-phase desorption kinetics and threshold thermal effects.

Conclusions:

• Established a robust and interpretable ML framework for optimizing PAHs thermal desorption.
• Demonstrated improved prediction accuracy and generalization through advanced ML techniques and data transformations.
• Provided practical insights for enhancing remediation reliability, energy efficiency, and intelligent process control in thermal desorption.