PM2.5 and O3 concentration estimation based on interpretable machine learning

High concentrations of PM2.5 and ozone (O3) seriously threaten human health. In this study, we constructed a machine learning-based model to predict PM2.5 and O3 concentrations, with the Fenwei Plain in China, as our study target. We evaluated the performance of RF, XGB, and CatBoost models for predicting PM2.5 and O3 concentrations and found that the CatBoost model performed the best, capturing most of the PM2.5 and O3 concentration changes (with R2 values above 0.7). Using the SHAP-based machine learning interpretation method, we analyzed the factors contributing to the variation of PM2.5 and O3 concentrations. We found that PM10 had the largest effect on PM2.5 concentration, while net surface solar radiation significantly affected O3 concentration. Additionally, meteorological variables such as temperature and humidity play a role in the variation of PM2.5 and O3 concentrations. Finally, based on the results of the SHAP analysis, we propose some feasible management methods for PM2.5 and O3, providing suggestions for managing PM2.5 and O3 concentrations.