Advanced Journal of Environmental Science and Technology

ISSN 2756-3251

Advances in Environmental Science and Technology | ISSN 2756-3278 | Vol. 17 (4), pp. 012-023, April, 2026

Research Article

Machine Learning-Driven Air Quality Forecasting and Emission Source Apportionment in Rapidly Urbanizing Megacities

Chidinma Eze¹*, Mateo Silva², Sora Takahashi³

¹Department of Environmental Management, University of Lagos, Lagos, Nigeria
²Department of Atmospheric Sciences, University of Chile, Santiago, Chile
³Department of Urban Environmental Engineering, Kyoto University, Kyoto, Japan

Corresponding Author: Chidinma Eze — [email protected]

Abstract

Rapid urbanization and industrial expansion have intensified fine particulate matter (PM2.5) and tropospheric ozone pollution across developing and emerging megacities. This study develops a hybrid deep-learning framework combining graph neural networks with positive matrix factorization to forecast multi-pollutant concentrations and apportion emission sources at neighborhood resolution. Using three years of high-frequency sensor data, meteorological reanalyses, and satellite aerosol optical depth records from Lagos, Santiago, and Tokyo, the model was evaluated against conventional chemical transport simulations. The hybrid architecture improved 48-hour PM2.5 prediction accuracy by 28% while accurately isolating contributions from vehicular exhaust, biomass burning, construction dust, and industrial plumes. Spatial vulnerability mapping revealed that low-income transit corridors experience disproportionate peak exposures during stagnant atmospheric boundary layer events. These findings demonstrate that integrating low-cost Internet-of-Things sensor networks with interpretable machine learning provides municipal authorities with a cost-effective decision-support tool for targeted emission controls and real-time public health advisories.

Keywords: air quality forecasting; source apportionment; machine learning; urban air pollution; fine particulate matter; environmental sensing

Manuscript Timeline: Received 05 February, 2026; Revised 14 March, 2026; Accepted 04 April, 2026; Published 12 April, 2026.

Citation: Eze, C., Silva, M., and Takahashi, S. (2026). Machine Learning-Driven Air Quality Forecasting and Emission Source Apportionment in Rapidly Urbanizing Megacities. Advances in Environmental Science and Technology, 17(4), pp. 012-023.