Related Experiment Video
Updated: Aug 27, 2025
13:00
Measuring Attention and Visual Processing Speed by Model-based Analysis of Temporal-order Judgments
Published on: January 23, 2017
10.0K
A Visual Analytics System for Improving Attention-based Traffic Forecasting Models
IEEE Transactions on Visualization and Computer Graphics
|September 26, 2022
Summary
This study introduces AttnAnalyzer, a visual analytics system for exploring deep learning (DL) traffic forecasting models. It helps analyze complex spatio-temporal dependencies to improve model performance.
Area of Science:
- Traffic forecasting
- Visual analytics
- Deep learning models
Background:
- Deep learning (DL) models excel in various tasks, including traffic forecasting.
- Analyzing DL models for traffic prediction is challenging due to their black-box nature and complex spatio-temporal data dependencies.
Purpose of the Study:
- To design a visual analytics system, AttnAnalyzer, for exploring DL model predictions in traffic forecasting.
- To enable effective spatio-temporal dependency analysis for better understanding of DL model behavior.
Main Methods:
- Developed AttnAnalyzer, a visual analytics system integrating dynamic time warping (DTW) and Granger causality tests.
- Incorporated multiple views (map, table, line chart, pixel) for dependency and model behavior analysis.
Main Results:
- AttnAnalyzer facilitates exploration of DL model behaviors in traffic forecasting.
- Case studies demonstrate improved model performance in two distinct road networks using the system.
Conclusions:
- AttnAnalyzer provides a valuable tool for domain experts to analyze and enhance DL traffic forecasting models.
- Effective spatio-temporal dependency analysis is crucial for understanding and improving complex DL models in the traffic domain.
