Predicting the risk of GenX contamination in private well water using a machine-learned Bayesian network model

Area of Science:

• Environmental Chemistry and Toxicology
• Environmental Engineering
• Geospatial Data Science

Background:

• Per- and polyfluoroalkyl substances (PFAS) are persistent environmental contaminants with complex characteristics challenging traditional modeling.
• Accurate prediction of PFAS spatial distribution is crucial for environmental risk assessment and public health protection.

Purpose of the Study:

• To develop and validate a machine learning model for predicting the risk of GenX, a PFAS compound, exceeding health goals in private drinking water wells.
• To identify key environmental factors influencing GenX contamination in private well water near a fluorochemical manufacturing facility.

Main Methods:

• Integrated spatial data from 1207 private wells with a mechanistic air deposition model and environmental datasets (soil, land use, topography, weather, source proximity).
• Trained a Bayesian network model using linked data to predict the probability of GenX exceeding a state provisional health goal (140 ng/L).
• Validated model accuracy using five-fold cross-validation, achieving a high ROC curve index (0.85).

Main Results:

• The Bayesian network model demonstrated high accuracy in predicting GenX risk in private wells.
• Historic atmospheric deposition rate of GenX from the manufacturing facility was the most significant predictor of well water contamination.
• Generated spatial risk predictions to guide environmental investigations and public health interventions.

Conclusions:

• Machine learning, specifically Bayesian networks, offers a powerful approach for predicting PFAS contamination in complex environmental settings.
• Understanding historical deposition patterns is critical for assessing and mitigating PFAS risks in drinking water sources.
• The developed model supports targeted risk assessment and public health strategies for PFAS exposure.