Efficient Group-Aware Graph Neural Network for Air Quality Forecasting in Small-scale Spaces

Air pollution poses significant risks to life and health in small-scale spaces. Accurate air quality predictions can provide critical data hours or days in advance, facilitating preventive actions. Previous research has primarily concentrated on large areas, such as countries, with ample data, often overlooking smaller, data-limited spaces. In this paper, we present a novel hierarchical model designed for forecasting air quality in such conditions, an Efficient Group-Aware Graph Neural Network for Air Quality Forecasting in Small-scale Spaces (EGAGNN). This model has the potential to significantly improve our understanding of air quality dynamics in small, data-limited spaces with its ability to capture direct and indirect influences among nodes. It identifies potential node connections and forms groups using a differentiable grouping network. Additionally, the model employs an encoder to capture group correlations and applies a messagepassing mechanism to learn the final node representations. Extensive experiences demonstrate that our model outperforms existing approaches.