Abstract
Event detection on real-time traffic data is crucial for transportation systems, enhancing safety in autonomous systems and the resilience of urban mobility infrastructure. However, the complex, heterogeneous traffic stream data and spatial-temporal patterns pose a significant challenge to existing methods, which often fail to capture higher-order interactions among nodes and edges and semantic context. This paper proposes a generative, self-attention model layered with two stacked content-based layers based on a framework that incrementally handles the process of traffic stream networks. Experimental analysis on benchmark event -detection dataset demonstrates the effectiveness of the proposed methodology, attaining peak performance values of 0.80 in Normalized mutual information(NMI), 0.73 IN Adjusted Mutual Information (AMI), and 0.72 in Adjusted Rand Index(ARI) across the incremental block of heterogeneous data, the proposed method consistently outperforms state-of-the-art baselines.
| Original language | English |
|---|---|
| Article number | 5999 |
| Journal | Sensors |
| Volume | 26 |
| Issue number | 19 |
| DOIs | |
| Publication status | Published - 22 Sept 2026 |
UN SDGs
This output contributes to the following UN Sustainable Development Goals (SDGs)
-
SDG 11 Sustainable Cities and Communities
Keywords
- data stream
- graph neural network
- higher information network
- incremental learning
- meta-path
Cite this
- APA
- Author
- BIBTEX
- Harvard
- Standard
- RIS
- Vancouver