Predicting high exposure of per- and polyfluoroalkyl substances (PFASs) in Korean adults using machine learning

Per- and polyfluoroalkyl substances (PFASs) are synthetic chemicals with high persistence in the environment and biological systems, raising significant public health concerns. This study developed and evaluated machine learning (ML) models to predict high PFAS exposure among participants in the fourth cycle of the Korean National Environmental Health Survey (KoNEHS). The study assessed the effectiveness of a Full model incorporating 64 variables and a Compact model incorporating 10 basic health metrics using six different ML algorithms: random forest (RF), gradient boosting machine (GBM), eXtreme gradient boosting (XGBoost), logistic regression (LR), support vector machine (SVM), and K-nearest neighbors (KNN). The study revealed a robust performance across all algorithms, achieving a balanced accuracy of 82 %. Age was the most significant predictor, particularly in individuals aged >45 years. Other critical predictors included serum mercury and serum lead levels in the Full model and hemoglobin levels, alanine aminotransferase, red blood cell counts, and platelet counts in the Compact model. Hence, these findings underscore the potential of ML in improving the identification and management of populations at risk for high PFAS exposure. Consequently, even minimal data sets can yield high predictive accuracy, as demonstrated by the Compact model.