Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

TrafPS: A Shapley-based Visual Analytics Approach to Interpret Traffic

Zezheng Feng, Yifan Jiang, Hongjun Wang, Zipei Fan, Yuxin Ma, Shuang-Hua Yang, Huamin Qu, Xuan Song

TL;DR

This work tackles the interpretability gap in DL-based traffic prediction by introducing TrafPS, a visual analytics system that employs Shapley-based region SHAP and trajectory SHAP to quantify the influence of surrounding regions and individual trajectories on predicted traffic. It combines a ST-ResNet predictor on a $38\times36$ grid with an interpretation layer and three coordinated views (Map-Trajectory, Radar Glyph, Fine-grained Grid) to support multi-level exploration from macro region-to-region effects to micro grid-to-trajectory contributions. The approach is validated on real taxi trajectory data from Chengdu, with two case studies demonstrating how TrafPS identifies key routes and informs urban planning decisions, supplemented by expert interviews attesting to feasibility and usefulness. The paper also discusses parameter choices (e.g., clustering into 21 regions) and model generality across different prediction architectures, and outlines directions for real-time data integration and richer decision recommendations.

Abstract

Recent achievements in deep learning (DL) have shown its potential for predicting traffic flows. Such predictions are beneficial for understanding the situation and making decisions in traffic control. However, most state-of-the-art DL models are considered "black boxes" with little to no transparency for end users with respect to the underlying mechanisms. Some previous work tried to "open the black boxes" and increase the interpretability of how predictions are generated. However, it still remains challenging to handle complex models on large-scale spatio-temporal data and discover salient spatial and temporal patterns that significantly influence traffic flows. To overcome the challenges, we present TrafPS, a visual analytics approach for interpreting traffic prediction outcomes to support decision-making in traffic management and urban planning. The measurements, region SHAP and trajectory SHAP, are proposed to quantify the impact of flow patterns on urban traffic at different levels. Based on the task requirement from the domain experts, we employ an interactive visual interface for multi-aspect exploration and analysis of significant flow patterns. Two real-world case studies demonstrate the effectiveness of TrafPS in identifying key routes and decision-making support for urban planning.

TrafPS: A Shapley-based Visual Analytics Approach to Interpret Traffic

TL;DR

This work tackles the interpretability gap in DL-based traffic prediction by introducing TrafPS, a visual analytics system that employs Shapley-based region SHAP and trajectory SHAP to quantify the influence of surrounding regions and individual trajectories on predicted traffic. It combines a ST-ResNet predictor on a grid with an interpretation layer and three coordinated views (Map-Trajectory, Radar Glyph, Fine-grained Grid) to support multi-level exploration from macro region-to-region effects to micro grid-to-trajectory contributions. The approach is validated on real taxi trajectory data from Chengdu, with two case studies demonstrating how TrafPS identifies key routes and informs urban planning decisions, supplemented by expert interviews attesting to feasibility and usefulness. The paper also discusses parameter choices (e.g., clustering into 21 regions) and model generality across different prediction architectures, and outlines directions for real-time data integration and richer decision recommendations.

Abstract

Recent achievements in deep learning (DL) have shown its potential for predicting traffic flows. Such predictions are beneficial for understanding the situation and making decisions in traffic control. However, most state-of-the-art DL models are considered "black boxes" with little to no transparency for end users with respect to the underlying mechanisms. Some previous work tried to "open the black boxes" and increase the interpretability of how predictions are generated. However, it still remains challenging to handle complex models on large-scale spatio-temporal data and discover salient spatial and temporal patterns that significantly influence traffic flows. To overcome the challenges, we present TrafPS, a visual analytics approach for interpreting traffic prediction outcomes to support decision-making in traffic management and urban planning. The measurements, region SHAP and trajectory SHAP, are proposed to quantify the impact of flow patterns on urban traffic at different levels. Based on the task requirement from the domain experts, we employ an interactive visual interface for multi-aspect exploration and analysis of significant flow patterns. Two real-world case studies demonstrate the effectiveness of TrafPS in identifying key routes and decision-making support for urban planning.
Paper Structure (26 sections, 1 theorem, 17 equations, 8 figures, 2 tables)

This paper contains 26 sections, 1 theorem, 17 equations, 8 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Visual Analytics for Traffic Data
  4. Visual Interpretation for Urban Traffic
  5. Explainable AI Approach for Traffic Prediction
  6. Task Analysis and Solution Overview
  7. Task Analysis
  8. Solution Overview
  9. Model
  10. Data Processing
  11. Prediction-Interpretation Layer
  12. Prediction Model
  13. Shapley Value
  14. Region SHAP
  15. Trajectory SHAP
  16. ...and 11 more sections

Key Result

Theorem 4.1

Given a region explanation saliency maps $E(g^{c,i,j}_t)$, the contribution of each trajectory in $g^{i, j}$ is $E(g^{c,i,j}_t \rightarrow t_r) =\frac{1}{\mid C^{c,i,j} \mid} E(g^{c,i,j}_t)$, where the subset trajectories $C$ is defined as where $N$ is the length of trajectory $t_r$, and $C^{in,i,j}$ and $C^{out,i,j}$ is to collect the trajectories where flow in or flow out the region $g^{i, j}$

Figures (8)

  • Figure 1: The overview of TrafPS. TrafPS consists of three parts, including the (A) data processing phase, (B) prediction interpretation layer, and (C) visual interface. The input of TrafPS is vehicle trajectory data and road network data.
  • Figure 2: The design rationale consists of three stages, including (A) overview stage, (B) exploration stage, and (C) fine-grain stage.
  • Figure 3: The design of map-trajectory view. The map-trajectory view consists of three parts, including ($a_1$) a dashboard, ($a_2$) a geography map, and (B) a traffic flow prediction heatmap.
  • Figure 4: The results of regional aggregation. (A) is the result of the division of the Voronoi Diagram for the urban area, and (B) is the visual effect of the border of each cluster.
  • Figure 5: The (A) design and the (B) example of radar glyph.
  • ...and 3 more figures

Theorems & Definitions (1)

  • Theorem 4.1