From micro to macro: multi-scale causal emergent complexity analysis in traffic dynamics

Accurate traffic prediction is essential for modern urban transportation systems. In this paper, we propose a novel multi-scale causal modeling framework for traffic prediction, leveraging Emergent Complexity (EC 2. 0) analysis to uncover the hierarchical nature of traffic dynamics. Through detailed experiments on real-world datasets (PEMS08, PEMS04, METR-EC) and Beijing Freeway & Rural Scenario datasets, we demonstrate the dominance of spatial dependencies in traffic patterns while highlighting the complementary role of temporal relationships. Our approach identifies critical nodes and paths within the network, revealing congestion-prone areas and informing targeted interventions. Furthermore, we show that our multi-scale model outperforms existing methods, achieving state-of-the-art results across all datasets. By analyzing phase transitions in causal gains (₂₏) and training dynamics, we provide interpretable insights into the model’s learning process and causal structure discovery. This work advances the field of traffic prediction by integrating causal reasoning and multi-scale modeling, offering both high accuracy and interpretability.