Particulate matter pollution prone-areas mapping using an enhanced explainable ensemble meta-algorithm

Particulate matter pollution in the air, especially PM_2.5 and PM₁₀, remains a critical environmental concern, particularly in urban areas like Tehran. Identifying areas susceptible to particulate matter pollution through mapping is crucial for effective public health management, informed urban planning, and formulating policies to combat air pollution. This study introduces an advanced approach to spatial modeling and susceptibility mapping of PM_2.5 and PM₁₀ pollutants in Tehran, utilizing an advanced XGBoost (eXtreme Gradient Boosting) algorithm and metaheuristic optimization algorithms (Bat and artificial bee colony (ABC)). Moreover, due to the black-box nature of machine learning algorithms in this research, an interpretable approach, namely SHAP (Shapley Additive exPlanations), has been employed. Significantly, higher temperatures, lower altitudes, moderate rainfall, and proximity to industrial zones were identified as crucial factors by the SHAP method. The evaluation results of particulate matter susceptibility maps using the receiver operating characteristic (ROC) curve indicated a higher accuracy for the XGBoost-Bat (PM_2.5 = 95.9% and PM₁₀ = 97.4%) algorithm than the XGBoost-ABC (PM_2.5 = 95.5% and PM₁₀ = 96.7%) and XGBoost (PM_2.5 = 92.8% and PM₁₀ = 96.3%) algorithms. This integrated framework provides a comprehensive understanding of particulate matter pollution and offers valuable insights for targeted interventions and urban planning in Tehran.