SMA-Hyper: Spatiotemporal Multi-View Fusion Hypergraph Learning for Traffic Accident Prediction

TL;DR

This work tackles the challenge of predicting traffic accidents under sparse, high-dimensional urban data by proposing SMA-Hyper, a Spatiotemporal Multiview Adaptive HyperGraph Learning framework. It fuses four views of urban information through adaptive pairwise graphs and setwise hypergraphs, leverages a multiview spatiotemporal encoder with gated temporal convolutions and attention fusion, and decodes predictions while incorporating dynamic external features. A local-global contrastive loss aligns linearly and nonlinearly related representations to mitigate sparsity and enhance generalization. Empirical evaluation on the London STATS19 dataset demonstrates substantial improvements in RMSE, MAE, and Recall over strong baselines, with ablation studies confirming the value of PKDE, hypergraph learning, and multiview fusion for capturing high-order spatiotemporal dependencies. The framework offers a scalable, interpretable approach to urban traffic safety, capable of integrating diverse city data and adapting to evolving urban configurations.

Abstract

Predicting traffic accidents is the key to sustainable city management, which requires effective address of the dynamic and complex spatiotemporal characteristics of cities. Current data-driven models often struggle with data sparsity and typically overlook the integration of diverse urban data sources and the high-order dependencies within them. Additionally, they frequently rely on predefined topologies or weights, limiting their adaptability in spatiotemporal predictions. To address these issues, we introduce the Spatiotemporal Multiview Adaptive HyperGraph Learning (SMA-Hyper) model, a dynamic deep learning framework designed for traffic accident prediction. Building on previous research, this innovative model incorporates dual adaptive spatiotemporal graph learning mechanisms that enable high-order cross-regional learning through hypergraphs and dynamic adaptation to evolving urban data. It also utilises contrastive learning to enhance global and local data representations in sparse datasets and employs an advance attention mechanism to fuse multiple views of accident data and urban functional features, thereby enriching the contextual understanding of risk factors. Extensive testing on the London traffic accident dataset demonstrates that the SMA-Hyper model significantly outperforms baseline models across various temporal horizons and multistep outputs, affirming the effectiveness of its multiview fusion and adaptive learning strategies. The interpretability of the results further underscores its potential to improve urban traffic management and safety by leveraging complex spatiotemporal urban data, offering a scalable framework adaptable to diverse urban environments.

Figures (8)

• Figure 1: Illustration of complex geographic relationships: Regions 1, 2, and 3 share similar POI distributions, yet only Regions 1 and 2 are directly linked by spatiotemporal correlations.
• Figure 2: The distribution of accident risk ratios in different conditions. The weather conditions include five distinct scenarios, while the greenspace ratio is divided into ten discrete clusters.
• Figure 7: Overview of the Spatiotemporal Multiview Adaptive HyperGraph Learning (SMA-Hyper) Framework. Section (a) depicts the graph construction process, integrating historical accident data with urban features into the graph and hypergraph structures. Section (b) presents the encoder phase, incorporating contrastive learning to enhance feature representation and differentiation. Section (c) illustrates the multi-view fusion decoder, which synthesizes various urban data streams like POIs and road information, culminating in the predictive output for traffic accident risks
• Figure 8: Spatial and Temporal Comparison of traffic accident data visualizations of London, 2021.
• Figure 9: Step-wise performance on all evaluation metrics for different models.