Single-particle decoding of aerosol pollutants size-composition relationships: An interpretable XGBoost-SHAP...

Constrained by the spatial resolution limitations of conventional analytical techniques and the inherent complexities of mass spectrometry data current understanding of aerosol pollutants behavior predominantly relies on statistical correlations. Addressing the fundamental challenge of resolving particle size-chemical composition relationships at the single-particle level, this study proposes an innovative interpretable framework integrating Dual-Threshold Entropy-Weighted Denoising with eXtreme Gradient Boosting-SHapley Additive exPlanations (DTEWD-XGBoost-SHAP). This framework is based on single particle aerosol mass spectrometry (SPAMS) data and uses DTEWD to effectively improve mass spectrometry quality. Subsequently, XGBoost and SHAP were combined to reveal for the first time the particle size regulation mechanism of key chemical components. To verify the feasibility of the DTEWD-XGBoost-SHAP framework, this study analyzed complex soldering smoke. After DTEWD processing, the prediction accuracy (R²) of single-particle aerosol particle size reached 0.759, and the root mean square error (RMSE) of the test set was reduced by 36.5 %. Based on SHAP analysis, it was found that carbon-containing fragments ions such as C₅⁺ significantly contribute to the growth of particle size through the enrichment effect, while anions such as NO₃^- and Cl^- indirectly regulate the distribution of the particle size through adsorption. The framework further validated its generalization ability in the soldering temperature classification task, achieving an accuracy rate of 93.75 % on the test set, and combined SHAP to analyze the impact of key components. This work advances mechanistic understanding of industrial aerosol composition-size relationships, enriches the methodological toolkit for studying aerosol formation and evolution dynamics, and pioneers novel analytical perspectives for deep mass spectrometry mining.