Ai for a breathable future: Predictive insights, health benefits & green abatement strategies

Abstract

The rapid expansion of urban areas has intensified challenges related to air pollution, posing significant risks to public health, ecological stability, and urban resilience. Current forecasting and mitigation strategies often face high computational costs, limited scalability, and insufficient integration of health impacts, creating a gap between theoretical models and their practical application in real-world smart city contexts. For instance, a highly sophisticated deep learning model can accurately predict air pollution levels in controlled settings. However, its deployment within a real-time smart city infrastructure can prove impractical due to the demands for high computational power, expensive hardware, and prolonged processing times. Consequently, city administrators cannot use these predictions quickly enough to issue a timely health advisory or implement mitigation strategies. To address these challenges, this dissertation presents a comprehensive, energy-efficient, AI-driven framework for urban air quality management that integrates real-time forecasting, sustainable mitigation strategies, and health impact assessments. The framework features a multi-layered architecture that combines convolutional and recurrent neural networks with federated optimization and edge computing. This design ensures accurate, low-latency predictions while minimizing computational demands. A spatiotemporal deep learning model captures local dependencies and long-term temporal dynamics, enabling robust air quality forecasting across diverse urban landscapes. Moreover, federated learning techniques are employed to safeguard data privacy and enhance convergence on heterogeneous datasets, while lightweight, energy-efficient models incorporate environmental and industrial parameters to facilitate low-power deployment. Evaluation using extensive urban datasets demonstrates that the proposed models/frameworks achieve superior prediction accuracy, faster convergence, and reduced energy consumption compared to existing methods. In addition, a health-oriented analysis highlights the correlation between pollution exposure, demographic factors, and public health risks. At the same time, the integration of meteorological data reveals strong associations between heatwave events and declining air quality. These findings offer actionable insights for emission control, sustainable urban planning, and proactive health risk management. By aligning predictive intelligence with national smart city and green growth strategies, the framework presents scalable, adaptive solutions that promote healthier and more resilient urban ecosystems. It equips policymakers, planners, and public health authorities with decision-support tools designed to enhance sustainability, livability, and overall human well-being.