Clustering and machine learning techniques identify air pollution regimes in Greater Cairo

Abstract

Air pollution is a major environmental and public health challenge in Greater Cairo due to rapid urbanization, intense traffic activity, and recurrent regional dust intrusions. This study presents a data driven analytical framework for identifying and interpreting recurring air pollution regimes in the city. The framework combines unsupervised K-means clustering with supervised Decision Tree (DT) and Random Forest (RF) models using atmospheric reanalysis data derived from the Copernicus Atmosphere Monitoring Service (CAMS) for the period 2023-2024. The optimized K-means model identified four distinct pollution regimes. Low pollution conditions accounted for approximately 75.1% of the analyzed period, whereas higher pollution regimes were mainly associated with traffic related emissions and episodic dust events. To assess the separability of the identified regimes, DT and RF classifiers were trained to predict cluster membership. The optimized Decision Tree achieved an accuracy of 93.10%, while the Random Forest model showed better classification performance, reaching a maximum accuracy of 97.49%, with a practical optimum of 97.43% obtained using 300 trees. Feature importance analysis showed that NO₂ was the dominant variable for distinguishing traffic related pollution regimes, whereas PM₁₀ played a key role in identifying dust related events. Overall, the findings indicate that integrating clustering with tree based classification provides an interpretable and effective approach for characterizing urban air pollution patterns. The resulting framework may support regime based air quality interpretation and targeted management strategies in Greater Cairo.