Predictive machine learning and geospatial modeling reveal PM10 hotspots and guide targeted air pollution interventions in Addis Ababa, Ethiopia

Air pollution is a critical twenty-first century environmental and public health challenge that is linked to millions of deaths and ecological harm. Accurate prediction of pollutants such as PM10 is vital for mitigation and urban sustainability. This study combines geospatial modeling with three machine learning algorithms (MLAs), Random Forest (RF), K-Nearest Neighbor (KNN), and Naïve Bayes (NB), to identify PM10 hotspots in Addis Ababa, Ethiopia. PM10 data from 11 stations (August 2021–August 2023) were analyzed alongside 25 atmospheric, climatic, anthropogenic, and pollution source predictors. A concentric zonal approach was used to assess spatial variability across radial distances and directional sectors and was supported by 30 m-resolution satellite imagery, climate data, and local geospatial repositories. The model accuracies were 0.95 (KNN), 0.93 (RF), and 0.88 (NB), with distinct performance trade-offs: RF predicted the largest “Good” PM10 zones (78.98 km2), KNN highlighted the most “UnHealSen” areas (279 km2), and NB predict “Moderate” coverage (311 km2). High PM10 concentrations clustered in eastern and northwestern sectors, aligning with industrial zones and traffic density. The results demonstrate the efficacy of MLAs and geospatial integration in producing high-resolution pollution maps. We advocate for targeted emission controls in hotspots, expanding public transit to reduce vehicular emissions, and incorporating air quality metrics into urban planning. This study advances air quality assessment methods for rapidly urbanizing regions, providing data-driven strategies to combat pollution and enhance ecological resilience in African cities.