Dynamic analysis for PFAS removal from contaminated water by foam separation based on machine learning models

Area of Science:

• Environmental Science
• Chemical Engineering
• Data Science

Background:

• Per- and polyfluoroalkyl substances (PFAS) are persistent global environmental contaminants with known toxicity.
• Foam separation (FS) is a promising, cost-effective remediation technique for PFAS removal.
• FS efficiency is highly variable due to complex interactions between parameters.

Purpose of the Study:

• To develop a predictive model for optimizing foam separation efficiency in PFAS remediation.
• To identify key factors influencing PFAS removal using machine learning.
• To establish a data-driven framework for enhancing FS performance.

Main Methods:

• Evaluation of six machine learning (ML) models for predicting PFAS removal efficiency.
• Development of an improved extreme gradient boosting (XGB) model, termed PFAS-XGB.
• Analysis of model interpretation and causal contribution of various factors.

Main Results:

• Extreme gradient boosting (XGB) demonstrated the highest predictive performance (R²).
• The PFAS-XGB model streamlined input variables and enhanced prediction efficiency.
• Aeration time and PFAS molecular weight were identified as critical determinants of removal efficiency.
• Operational conditions (56.8%) had the most significant impact, followed by PFAS properties (21.9%), surfactants (16.8%), and metal activators (4.5%).

Conclusions:

• Machine learning provides a robust framework for optimizing foam separation in PFAS remediation.
• Operational parameters are the primary drivers of FS efficiency, but surfactant and metal activator impacts are significant.
• The study offers actionable insights for improving PFAS removal strategies through data-driven optimization.