Root Cause Analysis of Traffic Anomalies Using Uneven Diffusion Model

Guang-Li Huang, Ke Deng, Yongli Ren, Jianxin LiIEEE Access 2019 · OpenReview

Abstract

Detection and analysis of traffic anomalies are important for the development of intelligent transportation systems. In particular, the root causes of traffic anomalies in road networks as well as their propagation and influence to the surrounding areas are highly meaningful. The root cause analysis of traffic anomalies aims to identify those road segments, where the traffic anomalies are detected by the traffic statuses significantly deviating from the usual condition and are originated due to incidents occurring in those roads such as traffic accidents or social events. The existing methods for traffic anomaly root cause analysis detect all traffic anomalies first and then apply, implicitly or explicitly, specified causal propagation rules to infer the root cause. However, these methods require reliable detection techniques to accurately identify all traffic anomalies and extensive domain knowledge of city traffic to specify plausible causal propagation rules in road networks. In contrast, this paper proposes an innovative and integrated root cause analysis method. The proposed method is featured by 1) defining a visible outlier index as the probabilistic indicator of traffic anomalies/disturbances and 2) automatically learning spatiotemporal causal relationship from historical data to build an uneven diffusion model for root cause analysis. The accuracy and effectiveness of the proposed method have been demonstrated by experiments conducted on a trajectory dataset with 2.5 billion location records of 27 266 taxies in Shenzhen city.

Introduction

Traffic anomalies, such as sudden congestion caused by accidents or social events, do not remain isolated; they ripple through road networks over time. Identifying the "root cause"—the specific road segment where the trouble originated—is vital for intelligent transportation systems to manage flow and clear incidents effectively.

Existing methods face two major hurdles: the "Threshold Trap," where binary detection ignores the probabilistic nature of traffic, and the "Heat Assumption," which treats traffic spread like uniform heat diffusion. In reality, traffic propagation is "uneven," depending on road topology, time of day, and complex driver behavior. This paper introduces an integrated framework to address these challenges by moving away from binary detection toward a probabilistic approach.

Methodology

The proposed solution consists of a two-pronged approach: a probabilistic indicator for anomalies and a deep-learning-based diffusion model.

Visible Outlier Index (VOI)

Instead of a simple "congested vs. not congested" label, the authors use the Kolmogorov-Smirnov (KS) test to compare the current distribution of vehicle speeds in a region against historical data. They calculate a Visible Outlier Index (VOI), which is a rescaled p-value representing the probability that the current traffic state is an anomaly. This allows the model to handle uncertainty and subtle deviations more gracefully than a hard threshold.

Uneven Diffusion Model

To understand how traffic spreads, the authors built a deep learning architecture using a Stacked AutoEncoder (SAE). The model is trained to predict the VOI of a target region based on its "spatiotemporal neighborhood."

Spatiotemporal neighborhood mapping for region A

As illustrated in the figure above, the model learns how traffic states at time $t-1$ influence a specific region at time $t$. By training on historical data, the model implicitly discovers the connectivity and flow constraints of the city without requiring manually defined rules.

Root Cause Identification

The identification of the root cause is achieved through diagnostic prediction error. The model predicts what the traffic should look like based on its neighbors.

Spatial neighborhood representation for region B

If the actual observed VOI in a region (such as region B in the grid above) is significantly higher than the predicted VOI, it indicates that the disturbance did not flow in from the outside—it must have originated right there. This "surprise" in the model's prediction is the signature of a root cause.

Experimental Results

The model was validated using a massive real-world trajectory dataset from Shenzhen, China, consisting of 2.5 billion location records from 27,266 taxis. The city was partitioned into a grid of 42 × 90 regions, with time sliced into 5-minute bins.

Key results include:

  • High Precision: In a test identifying 100 manually verified root causes, the Uneven Diffusion Model achieved 88% precision and an 83% F1-score, vastly outperforming traditional Heat Diffusion models (55% precision).
  • Real-world Validation: Using the 2014 Shenzhen Marathon as a case study, the model successfully identified the marathon's starting point and subsequent road closures as the root causes of surrounding traffic anomalies.

Conclusion

By treating traffic status as a distribution and learning topology implicitly, the proposed framework provides a robust method for identifying the origins of traffic disturbances. Future work may involve applying these concepts to Graph Convolutional Networks (GCNs) to better align with the natural graph structure of road networks and integrating social media data to provide semantic context for identified incidents.

Summary updated

Abstract

Detection and analysis of traffic anomalies are important for the development of intelligent transportation systems. In particular, the root causes of traffic anomalies in road networks as well as their propagation and influence to the surrounding areas are highly meaningful. The root cause analysis of traffic anomalies aims to identify those road segments, where the traffic anomalies are detected by the traffic statuses significantly deviating from the usual condition and are originated due to incidents occurring in those roads such as traffic accidents or social events. The existing methods for traffic anomaly root cause analysis detect all traffic anomalies first and then apply, implicitly or explicitly, specified causal propagation rules to infer the root cause. However, these methods require reliable detection techniques to accurately identify all traffic anomalies and extensive domain knowledge of city traffic to specify plausible causal propagation rules in road networks. In contrast, this paper proposes an innovative and integrated root cause analysis method. The proposed method is featured by 1) defining a visible outlier index as the probabilistic indicator of traffic anomalies/disturbances and 2) automatically learning spatiotemporal causal relationship from historical data to build an uneven diffusion model for root cause analysis. The accuracy and effectiveness of the proposed method have been demonstrated by experiments conducted on a trajectory dataset with 2.5 billion location records of 27 266 taxies in Shenzhen city.
BibTeX
@inproceedings{huang2019root,
  author = {Guang-Li Huang and Ke Deng and Yongli Ren and Jianxin Li},
  title = {{Root Cause Analysis of Traffic Anomalies Using Uneven Diffusion Model}},
  booktitle = {IEEE Access},
  year = {2019},
  url = {https://openreview.net/forum?id=elMaZZjDWd},
}
Figure