Assessing features of PFAS groundwater occurrence using SHAP-enhanced machine learning
Lin Wang1, Yun Ma1, Dulith Rajapakshe1
1Department of Civil and Environmental Engineering, New Mexico State University, Las Cruces, NM, 88003, United States.
Environmental Pollution (Barking, Essex : 1987)
|July 28, 2026
Summary
Explainable machine learning models accurately predict per- and polyfluoroalkyl substances (PFAS) in groundwater. Key predictors include sampling year and proximity to PFAS sources, offering a transparent framework for risk assessment.
Area of Science:
- Environmental Science
- Data Science
- Toxicology
Background:
- Per- and polyfluoroalkyl substances (PFAS) are persistent groundwater contaminants with significant public health implications.
- Predicting PFAS occurrence is complex due to high-dimensional environmental data and challenges in model interpretability.
Purpose of the Study:
- To benchmark explainable machine learning models for predicting PFAS occurrence in groundwater.
- To identify key features associated with estimated PFAS occurrence using interpretable models.
Main Methods:
- Utilized a comprehensive dataset of 12,406 groundwater records (2001-2019) with 172 features.
- Evaluated four tree-based ensemble classifiers (Random Forest, XGBoost, LightGBM, CatBoost) using binary classification.
- Employed SHapley Additive exPlanations (SHAP) for model interpretability.
Main Results:
- Binary classification achieved >94% testing accuracy, demonstrating robust and generalizable performance.
- Sampling year and proximity to major PFAS sources were the dominant predictors across all models.
- SHAP analysis revealed temporal variations in predictor importance, potentially linked to regulatory or monitoring changes.
Conclusions:
- Developed an interpretable, data-driven framework for predicting PFAS occurrence in groundwater.
- Identified key environmental and temporal factors influencing PFAS contamination.
- Provided a transparent approach to inform PFAS risk assessment and management strategies.
