Global atmospheric pollutant exposure is spatially heterogeneous and systematically linked to socioeconomic vulnerability, yet integrated assessments combining compositional data with population adaptive capacity remain scarce. This study analyzes the distribution of PM \(_{2.5}\) , NO \(_2\) , O \(_3\) , and CO across 12,438 cities in 195 countries, developing an Integrated Exposure Metric (IEM) that couples pollutant burden with socioeconomic vulnerability. Ensemble machine learning is applied as a consistency check to confirm which atmospheric components dominate the integrated burden. PM \(_{2.5}\) and O \(_3\) contribute near-equally to the global pollutant mixture (48% and 46%, respectively), yet their socioeconomic gradients diverge markedly: PM \(_{2.5}\) accounts for 57% of exposure burden in low-income regions versus 32% in high-income regions, while ozone displays the inverse gradient. Sensitivity analysis across three alternative CPB weighting scenarios confirms that PM \(_{2.5}\) explains 69–86% of IEM variance regardless of weight specification, demonstrating that this finding is not an artifact of pre-assigned pollutant weights. Clustering stability analysis identifies two compositionally distinct exposure regimes (silhouette = 0.686), with the higher-burden cluster of 6,551 cities concentrated in South Asia and the Middle East. Corrected Gini coefficients indicate that NO \(_2\) exhibits the most unequal spatial distribution (Gini = 0.696), followed by PM \(_{2.5}\) (0.607), while O \(_3\) is most spatially uniform (0.347). These findings indicate that effective atmospheric management requires source-specific strategies: primary PM \(_{2.5}\) controls for developing regions and precursor management for industrialized areas. This framework offers a quantitative foundation for equitable air quality interventions aligned with WHO guidelines and UN Sustainable Development Goals.
