Machine Learning–Enhanced PFAS Contaminant Fate and Transport Modeling for Optimized Groundwater Remediation Strategies at Industrial Sites

Abstract

This study investigates the persistent challenge of accurately modeling per and polyfluoroalkyl substances (PFAS) fate and transport in groundwater at industrial sites, where heterogeneous geology, variable groundwater flow, mixed PFAS chemistries, and uncertain source zones often weaken remediation planning and delay effective intervention. The purpose of the research was to examine whether machine learning enhanced PFAS fate and transport modeling can improve the optimization of groundwater remediation strategies in complex industrial settings. The study adopted a quantitative, cross sectional, case-based design and drew on a purposive sample of 120 professionals from cloud and enterprise style environmental decision contexts, including environmental engineers, hydrogeologists, remediation specialists, consultants, industrial site managers, and regulatory professionals. The key variables were PFAS transport complexity, machine learning modeling capability, machine learning predictive trust, PFAS predictive modeling quality, and optimized groundwater remediation strategies. Data were collected through a structured 5-point Likert scale questionnaire and analyzed using descriptive statistics, Cronbach’s alpha, Pearson correlation, and multiple regression in SPSS. Reliability was strong across all constructs, with Cronbach’s alpha values ranging from 0.82 to 0.89 and an overall instrument reliability of 0.86. Descriptive findings showed high mean scores for PFAS transport complexity (M = 4.21, SD = 0.61), machine learning modeling capability (M = 4.08, SD = 0.66), predictive trust (M = 3.94, SD = 0.70), predictive modeling quality (M = 4.12, SD = 0.63), and remediation optimization (M = 4.18, SD = 0.59). Correlation analysis revealed that predictive modeling quality had the strongest association with optimized remediation strategies (r = 0.76, p < .01). Regression results showed that the model explained 68.4% of the variance in remediation optimization (R² = 0.684, F = 49.87, p < .001), with predictive modeling quality emerging as the strongest predictor (β = 0.38, p < .001), followed by machine learning capability (β = 0.29, p = .001), PFAS transport complexity (β = 0.24, p = .003), and predictive trust (β = 0.21, p = .006). The study concludes that machine learning enhanced modeling offers a practically significant pathway for improving PFAS remediation decision quality, plume control, treatment prioritization, and cost-efficient groundwater cleanup planning at industrial sites.