Spatio-Temporal Data Mining with Information Integrity Protection: Graph Signal Based Air Quality Prediction

Due to industrial development, air pollution has become a persistent issue. Accurately predicting air quality is challenging due to complex spatiotemporal correlations within data. Previous researches utilize diverse modules to extract features separately from the temporal and spatial dimensions of data. While these approaches achieved promising results, they overlooked the integrity of data information. To address this, we propose a novel SpatioTemporal Air Quality forecasting Network, namely ST-AQNet, which avoids extracting spatial and temporal features individually but transforms spatiotemporal data into high-dimensional graph signals to make predictions. ST-AQNet involves only interactions between graph signals, which maintains the integrity of the data information while reducing complexity and computational overhead compared to other methods. Extensive experiments on five real-world datasets demonstrate the superior performance of our model beyond state-of-the-art methods, even on different spatiotemporal forecasting tasks.