ICST-DNET: An Interpretable Causal Spatio-Temporal Diffusion Network for Traffic Speed Prediction

TL;DR

ICST-DNET tackles traffic speed forecasting by explicitly modeling traffic diffusion through temporal and spatial causality, while delivering interpretable explanations and adaptability to speed fluctuations. The architecture combines three modules: STCL to learn causal diffusion, CGG to generate time- and space-based causal graphs for interpretability, and SFPR to recognize speed fluctuation patterns with time-filtered spatio-temporal encoding. Empirical results on METR-LA and Ningxia-YC show that ICST-DNET achieves superior accuracy and provides causal explanations, outperforming baselines across multiple horizons and scenarios. The work advances practical ITS deployment by offering both high predictive performance and transparent diffusion reasoning via causal graphs and matrices.

Abstract

Traffic speed prediction is significant for intelligent navigation and congestion alleviation. However, making accurate predictions is challenging due to three factors: 1) traffic diffusion, i.e., the spatial and temporal causality existing between the traffic conditions of multiple neighboring roads, 2) the poor interpretability of traffic data with complicated spatio-temporal correlations, and 3) the latent pattern of traffic speed fluctuations over time, such as morning and evening rush. Jointly considering these factors, in this paper, we present a novel architecture for traffic speed prediction, called Interpretable Causal Spatio-Temporal Diffusion Network (ICST-DNET). Specifically, ICST-DENT consists of three parts, namely the Spatio-Temporal Causality Learning (STCL), Causal Graph Generation (CGG), and Speed Fluctuation Pattern Recognition (SFPR) modules. First, to model the traffic diffusion within road networks, an STCL module is proposed to capture both the temporal causality on each individual road and the spatial causality in each road pair. The CGG module is then developed based on STCL to enhance the interpretability of the traffic diffusion procedure from the temporal and spatial perspectives. Specifically, a time causality matrix is generated to explain the temporal causality between each road's historical and future traffic conditions. For spatial causality, we utilize causal graphs to visualize the diffusion process in road pairs. Finally, to adapt to traffic speed fluctuations in different scenarios, we design a personalized SFPR module to select the historical timesteps with strong influences for learning the pattern of traffic speed fluctuations. Extensive experimental results prove that ICST-DNET can outperform all existing baselines, as evidenced by the higher prediction accuracy, ability to explain causality, and adaptability to different scenarios.

Figures (10)

• Figure 1: (a) The traffic diffusion between Road 1 and its neighboring roads from timestep t-P to t. Solid black lines indicate the road network; solid green lines represent the direct diffusion between Road 1 and its first-order neighbors; blue dashed lines indicate the indirect diffusion between Road 1 and its second-order neighbors. Differences in the gradient color of the lines indicate different influence levels. (b) Examples of traffic speed fluctuation over time. Roads 14 and 28 are neighboring roads and exhibit similar patterns.
• Figure 2: Overview of the proposed ICST-DNET. Left: SFPR module. $\mathcal{H}$ is the hidden state of the input. $\mathcal{H}_I^{E n c}$ is the hidden states selected by the time filtering array as the input of the ST-Encoding. $\mathcal{H S T}^{E n c}$ is the output of ST-Encoding. $\mathcal{H}_I^{D e c}$ is the input of the ST-Decoding, and $H S T^{D e c}$ denotes the production of the ST-Decoding. Right: STCL and CGG module. $X^i[: P]$ represents the time series of road $i$ in historical $P$ timesteps. $\gamma_j$ represents the learnable survival probability of the $j$th residual block.
• Figure 3: The Temporal Causality Layer. We employ a ResNet model incorporating Learnable Layer Relevance $\gamma_q$ concerning the spatial causality. The outputs of all the residuals are forwarded to the SCL layer.
• Figure 4: ST-Encoding. There are three steps: 1) select relevant spatio-temporal neighbors based on the time filtering array, 2) spatial and temporal attention is designed to extract dynamic spatio-temporal correlations, and 3) dynamic spatio-temporal properties are adaptively fused via ST-Fusion.
• Figure 5: Training error (left) and validation error (right) in the Ningxia-YC. MAE is regarded as an error measure function.