Evaluating regional and seasonal variations of PM2.5 and O3 across Malaysia using XGBoost and SHAP-based interpretability.

Evaluating regional and seasonal air pollution variability is essential for monitoring tropical environments. This study evaluated regional and seasonal variations in fine particulate matter (PM2.5) and ground-level ozone (O3) across five regions of Malaysia using daily atmospheric and meteorological observations from 2018 to 2023 from the Department of Environment (DOE), complemented by ERA5 reanalysis variables (boundary layer height, solar radiation, and cloud cover). Multiple Linear Regression (MLR) and Extreme Gradient Boosting (XGBoost) were compared, while Shapley Additive Explanations (SHAP) were applied to interpret driver importance and nonlinear relationships. Kruskal-Wallis and Dunn post hoc tests identified regional and seasonal differences in both pollutants. PM2.5 varied spatially with (χ2 = 3239.530, p < 0.001) and seasonally (χ2 = 477.837, p < 0.001). O3 varied geographically with (χ2 = 2173.607, p < 0.001) and seasonally (χ2 = 69.579, p < 0.001). XGBoost outperformed MLR, achieving R2 = 0.576 and RMSE = 4.87 μg m-3 for PM2.5, and R2 = 0.543 and RMSE = 3.4 ppb for O3, with skill varying by region and season and increasing to R2 = 0.85 when persistence was included. SHAP analysis showed that PM2.5 variability was associated with combustion-related tracers and meteorological factors, whereas O3 was more strongly associated with humidity and temperature, with humidity exerting greater influence during the Northeast Monsoon. These suggest that air pollution variability across Malaysia is governed by interactions between emission patterns and meteorological conditions, emphasizing the importance of season-specific evaluation, explainable machine learning, and regional assessment for policy development.