High-fidelity prediction of surfactant-enhanced oil remediation efficiency using a strictly curated dataset and a tabular foundation model

This study introduces a novel, high-fidelity modeling framework for surfactant-enhanced oil remediation (SEOR). Our hybrid strategy integrates a strict data curation pipeline—creating a complete-case, high-integrity dataset by avoiding imputation—with a state-of-the-art tabular foundation model, the Prior-data Fitted Network (TabPFN). Systematically benchmarked against 40 conventional machine learning algorithms, the TabPFN framework demonstrated a significant advance in predictive capability, achieving a coefficient of determination (R²) of 0.958 and a root mean squared error (RMSE) of 6.03. This represents a 21.2 % reduction in RMSE over the best-performing traditional model, Extra Trees Regressor (R²=0.932, RMSE=7.65). The unique strength of TabPFN lies in its pre-training, which allows it to deliver state-of-the-art accuracy on small scientific datasets without hyperparameter tuning. This work delivers a superior tool for understanding contaminant mobilization and establishes a robust, generalizable paradigm for modeling complex environmental systems.