Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Discovering the Precursors of Traffic Breakdowns Using Spatiotemporal Graph Attribution Networks

Zhaobin Mo, Xiangyi Liao, Dominik A. Karbowski, Yanbing Wang

TL;DR

The paper tackles identifying and explaining precursors to traffic breakdowns by marrying spatiotemporal graph neural networks with Shapley-value based explanations. It extends Shapley-based attribution to spatiotemporal graphs through a mask-based what-if subgraph generator and a ST-GCN predictor. Applied to the I-24 highway data, the approach reveals that lane adjacency and abrupt braking are major precursors and that influential neighbors are shaped by road topology rather than mere distance. This framework enhances interpretability of traffic forecasting and supports targeted congestion mitigation strategies.

Abstract

Understanding and predicting the precursors of traffic breakdowns is critical for improving road safety and traffic flow management. This paper presents a novel approach combining spatiotemporal graph neural networks (ST-GNNs) with Shapley values to identify and interpret traffic breakdown precursors. By extending Shapley explanation methods to a spatiotemporal setting, our proposed method bridges the gap between black-box neural network predictions and interpretable causes. We demonstrate the method on the Interstate-24 data, and identify that road topology and abrupt braking are major factors that lead to traffic breakdowns.

Discovering the Precursors of Traffic Breakdowns Using Spatiotemporal Graph Attribution Networks

TL;DR

The paper tackles identifying and explaining precursors to traffic breakdowns by marrying spatiotemporal graph neural networks with Shapley-value based explanations. It extends Shapley-based attribution to spatiotemporal graphs through a mask-based what-if subgraph generator and a ST-GCN predictor. Applied to the I-24 highway data, the approach reveals that lane adjacency and abrupt braking are major precursors and that influential neighbors are shaped by road topology rather than mere distance. This framework enhances interpretability of traffic forecasting and supports targeted congestion mitigation strategies.

Abstract

Understanding and predicting the precursors of traffic breakdowns is critical for improving road safety and traffic flow management. This paper presents a novel approach combining spatiotemporal graph neural networks (ST-GNNs) with Shapley values to identify and interpret traffic breakdown precursors. By extending Shapley explanation methods to a spatiotemporal setting, our proposed method bridges the gap between black-box neural network predictions and interpretable causes. We demonstrate the method on the Interstate-24 data, and identify that road topology and abrupt braking are major factors that lead to traffic breakdowns.

Paper Structure

This paper contains 8 sections, 3 equations, 4 figures.

Table of Contents

  1. Introduction
  2. Problem Statement
  3. Methodology
  4. Mask Generator
  5. Spatiotemporal Prediction Model
  6. Explanation Generator
  7. Experiment Results
  8. Conclusion

Figures (4)

  • Figure 1: Illustration of traffic breakdowns. A traffic breakdown contains phases of trigger&formation (A), propagation (B) and dissipation (C). Our goal is to discover the potential traffic breakdown precursors from region X, which is the downstream area antecedent to the breakdown trigger.
  • Figure 2: Flowchart of the proposed framework.
  • Figure 3: Comparison of true and predicted flow and density across four lanes for the I-24 dataset. The distance unit is $0.1$ miles and the time unit is $10$ seconds.
  • Figure 4: Spatial distributions of Shapley values among different neighbors for various SAGs. The two subfigures correspond to different investigated nodes, highlighted in red—Lane 2 in (a) and Lane 3 in (b), respectively.