Air pollution has emerged as a pressing challenge in rapidly urbanizing regions, demanding accurate and timely forecasting solutions to mitigate its adverse health impacts. This study presents a comprehensive AI-driven framework that integrates multisource environmental data—comprising satellite imagery, ground-level sensors, meteorological inputs, and mobile IoT devices—for enhanced air quality forecasting in urban settings. Leveraging advanced deep learning models, particularly LSTM and Transformer-based architectures, the system captures complex spatio-temporal patterns in pollutant behavior. A key innovation of this work lies in its data fusion strategy, which synchronizes heterogeneous data streams to improve prediction reliability. Furthermore, the model incorporates a health risk assessment module that translates pollutant forecasts into actionable health indicators based on population demographics and WHO-defined exposure thresholds. Experimental results across multiple urban zones demonstrate significant improvements in predictive accuracy when compared to traditional statistical models, with RMSE reductions exceeding 20%. The system also offers real-time responsiveness through edge-enabled deployment, ensuring low latency in high-density urban environments. By bridging the gap between environmental sensing and public health analytics, this work contributes to smarter urban planning, policy intervention, and personalized health alerts. The proposed approach not only advances the technological frontiers of air quality monitoring but also provides a scalable model for integration within smart city ecosystems.

Keywords: Air Quality Forecasting, Multisource Data Fusion, Health Risk Assessment, Smart City, Edge Computing