[Paper Review] AST-GCN: Attribute-Augmented Spatiotemporal Graph Convolutional Network for Traffic Forecasting

Jiawei Zhu, Chao Tao+5 more|arXiv (Cornell University)|Nov 22, 2020

Traffic Prediction and Management Techniques34 references17 citations

TL;DR

This paper proposes AST-GCN, a novel spatiotemporal graph convolutional network that integrates both static (e.g., POI distribution) and dynamic (e.g., weather) external attributes into traffic forecasting. By modeling these attributes as dynamic and static features and fusing them via an attribute-augmented unit, AST-GCN improves prediction accuracy and interpretability, outperforming baseline models across multiple datasets and forecasting horizons.

ABSTRACT

Traffic forecasting is a fundamental and challenging task in the field of intelligent transportation. Accurate forecasting not only depends on the historical traffic flow information but also needs to consider the influence of a variety of external factors, such as weather conditions and surrounding POI distribution. Recently, spatiotemporal models integrating graph convolutional networks and recurrent neural networks have become traffic forecasting research hotspots and have made significant progress. However, few works integrate external factors. Therefore, based on the assumption that introducing external factors can enhance the spatiotemporal accuracy in predicting traffic and improving interpretability, we propose an attribute-augmented spatiotemporal graph convolutional network (AST-GCN). We model the external factors as dynamic attributes and static attributes and design an attribute-augmented unit to encode and integrate those factors into the spatiotemporal graph convolution model. Experiments on real datasets show the effectiveness of considering external information on traffic forecasting tasks when compared to traditional traffic prediction methods. Moreover, under different attribute-augmented schemes and prediction horizon settings, the forecasting accuracy of the AST-GCN is higher than that of the baselines.

Motivation & Objective

• Address the limitation of traditional traffic forecasting models that fail to incorporate external factors affecting traffic states.
• Improve prediction accuracy by modeling both static (e.g., POI density) and dynamic (e.g., weather) attributes as part of the spatiotemporal learning process.
• Enhance model interpretability by explicitly encoding and fusing external attributes into the graph convolutional framework.
• Demonstrate robustness and generalization of the model under noisy data conditions through perturbation analysis.
• Validate the effectiveness of attribute-augmentation through ablation studies and visualization of prediction performance on turning points and peaks.

Proposed method

• Model the road network as a spatial graph where nodes represent road segments and edges represent spatial connectivity.
• Introduce an attribute-augmented unit that separately encodes static attributes (e.g., POI types and counts) and dynamic attributes (e.g., weather conditions) as additional input features.
• Fuse the attribute embeddings with the traffic flow features through learnable attention or concatenation mechanisms before feeding into the graph convolutional layers.
• Apply graph convolutional networks (GCNs) to capture spatial dependencies and gated recurrent units (GRUs) to model temporal dynamics in the augmented feature space.
• Use a multi-head attention mechanism to dynamically weigh the importance of different attributes during forecasting.
• Train the end-to-end model using mean absolute error (MAE) and mean squared error (MSE) loss functions on real-world traffic datasets.

Experimental results

Research questions

• RQ1 Can the integration of static and dynamic external attributes improve the accuracy of spatiotemporal traffic forecasting?
• RQ2 How do static attributes (e.g., POI distribution) and dynamic attributes (e.g., weather) individually and jointly affect prediction performance?
• RQ3 Does the AST-GCN model generalize well under noisy data conditions, indicating robustness?
• RQ4 How does the model perform in capturing short-term versus long-term traffic trends, especially at turning points and peaks?
• RQ5 To what extent do external attributes enhance the model’s ability to predict critical traffic events such as congestion or sudden speed drops?

Key findings

• AST-GCN achieves superior forecasting accuracy compared to baseline models across multiple datasets, with significant improvements in MAE and RMSE metrics.
• The combination of static and dynamic attributes yields the best performance, indicating that diverse external information enhances predictive capability.
• The model demonstrates robustness under Gaussian and Poisson noise, with minimal degradation in performance across varying noise levels.
• Long-term forecasting (e.g., 60-minute horizon) shows higher error than short-term forecasting (e.g., 15-minute horizon), indicating better capture of short-range dependencies.
• Visualization results confirm that attribute-augmented predictions are closer to ground truth at turning points and peaks, especially when both static and dynamic attributes are used.
• Ablation studies show that models enhanced with external attributes outperform those without, particularly in capturing abrupt changes in traffic speed.