Background: This study aims to explore the various risk factors associated with exposure to polycyclic aromatic hydrocarbons (PAHs) and benzene in Korean adult males (n = 2744), using data from the Korean National Environmental Health Survey (KoNEHS) conducted from 2015 to 2017. Methods: Isolation Forest, a machine learning algorithm specialized in anomaly detection, was employed to identify key variables influencing urinary biomarkers such as 1-Hydroxypyrene, 2-Hydroxynaphthalene and trans-Muconic acid. Results: The results revealed that age, smoking, alcohol consumption, proximity to roads, and grilled food consumption were significant predictors. Smoking emerged as the most influential factor across all biomarkers, highlighting its substantial impact on PAHs and benzene exposure. Comparative analysis demonstrated that Isolation Forest outperformed traditional models like Chi-squared Automatic Interaction Detection (CHAID), KNN (k-Nearest Neighbors), and Random Forest in detecting exposure-related anomalies, achieving an accuracy of 92%, a recall of 89%, a precision of 90%, an F-1 score of 89.5%, and an Area Under the Curve (AUC) of 0.93, which were approximately 5–10% higher than those achieved by the other models. Multiple regression analysis confirmed the statistical significance of these variables, with smoking showing the highest standardized beta values across all biomarkers, indicating its predominant influence. Conclusions: The study underscores the potential of machine learning in enhancing exposure assessment and suggests policy interventions targeting behavioral risk factors, particularly smoking cessation. Future research should consider longitudinal approaches and include additional variables for a comprehensive exposure evaluation.