ST-DCAN: Spatio-Temporal Dual Compression Attention Network for Traffic Prediction

With the rapid expansion of urban scale, the demand for precise traffic prediction is increasing. While existing traffic prediction models often incorporate complex components, their marginal benefits are diminishing. Additionally, these models struggle to extract generalized patterns among global nodes and segments, which impacts their performance. In response to these challenges, we introduce a transformer-based model, the Spatio-temporal Dual Compression Attention Network (ST-DCAN), which integrates a unified attention mechanism on input patches. Our approach reduces computational complexity from exponential to linear levels, enhances the extraction of both temporal and spatial information, and achieves state-of-the-art performance across three real-world traffic forecasting datasets with optimal computational efficiency.