AI-Driven Air Quality Forecasting Models for Urban Pollution Management: A Comparative Study of Machine Learning Algorithms in Developing Economies

Abstract

Air pollution in urban environments is a major environmental and health issue, especially in the developing economies where the current indicators of environmental degradation are high due to rapid industrialization, automotive emissions and failure to enforce regulations on vehicles and industries. Proper prediction of air quality indices (AQI) is critical in activating mitigation policies and guidance of policy measures. Although traditional forecasting methods are of value in their own way, they tend to not fully capture the nonlinear, complex and spatio/temporal processes of urban air pollution. This work discusses the use of modelling with the use of Artificial Intelligence (AI) and specifically the Machine Learning (ML) in the context of predicting air quality to enable smarter management of urban pollution. It takes a comparative framework to compare the efforts of four mainstream ML models, which are Random Forest (RF) Support Vector Machines (SVM) Long Short-Term Memory (LSTM), and Extreme Gradient Boosting (XGBoost) on real-time and past air quality data of chosen metropolitan cities, India, Bangladesh, and Nigeria. The used methodology combines measurements of the monitoring stations of the air pollution levels located in urban areas and meteorological data and pollutant concentrations (PM2.5, PM10, NO 2, SO 2, CO, O 3 ). Statistical measures of performance of the models include Coefficient of Determination (R 2 ), Root Mean Square Error (RMSE), and Mean Absolute Error (MAE). According to the preliminary findings it has been observed that LSTM presents the best results regarding the temporal dependencies, and Random forest presents the best results regarding high pollution cases. Volumetric visualizations and GIS-based overlaps are also provided to demonstrate the difference in hotspots of the pollution and the sensitivity of various algorithms to the variability of data in locals. The results support the possibility of AI-driven forecasting as an expandable and flexible agent of urban environmental planning as well as emergency response systems in resource-limited conditions. Such a comparative study does not only reveal the strengths and weaknesses of the algorithms but also offers a plan of how to implement the predictive AI models in the realm of public air quality government in developing economies.