Season and weather factors matter, but not enough: a machine learning-based study on predicting incremental lifetime cancer risk of polycyclic aromatic hydrocarbons

With the intensification of urbanization, air pollution has garnered global concern. This study aims to predict the incremental lifetime cancer risk (ILCR) of polycyclic aromatic hydrocarbons (PAHs) in atmospheric PM_2.5. Utilizing machine learning regression algorithms and data from six cities in Jiangsu Province in 2018, we established models to investigate the relationship between ILCR and various factors, with a special emphasis on seasonal and meteorological data. After model training, SHapley Additive exPlanation (SHAP) analysis revealed that seasonal factors were even more influential than PM_2.5 in predicting ILCR. Models were then validated using 2019 data, resulting in an R² of 0.42, which indicated a decrease in accuracy compared to the 2018 test set R² of 0.74 but still represented an improvement over using PM_2.5 alone (R² = 0.2). This suggests that while seasonal and related factors are crucial, additional factors are needed to build a robust model for future ILCR predictions.