Trffc: efficient traffic forecasting through adaptive spatio-temporal graph reduction

The complex interconnections in modern traffic networks pose significant challenges for storage, noise reduction, and interpretability, all of which are essential for accurate forecasting and analysis-key factors in effective traffic management and planning. Graph sparsification mitigates these challenges by reducing edges while preserving essential structures, enabling more efficient processing. However, existing sparsification algorithms are often static and fail to address the dynamic nature of evolving traffic networks, reflecting changes in traffic flow patterns. This paper introduces, Trffc, a reinforcement learning-based framework to sparsify temporal traffic graphs, enhancing both the scalability of forecasting and the recognition of emerging patterns. Through practical demonstrations, experiments on real-world temporal traffic data show that Trffc preserves key network properties and significantly boosts the performance of traffic forecasting models.