Jove
Visualize
Contact Us
JoVE
x logofacebook logolinkedin logoyoutube logo
ABOUT JoVE
OverviewLeadershipBlogJoVE Help Center
AUTHORS
Publishing ProcessEditorial BoardScope & PoliciesPeer ReviewFAQSubmit
LIBRARIANS
TestimonialsSubscriptionsAccessResourcesLibrary Advisory BoardFAQ
RESEARCH
JoVE JournalMethods CollectionsJoVE Encyclopedia of ExperimentsArchive
EDUCATION
JoVE CoreJoVE BusinessJoVE Science EducationJoVE Lab ManualFaculty Resource CenterFaculty Site
Terms & Conditions of Use
Privacy Policy
Policies

Related Experiment Videos

Interpretable spatiotemporal traffic crash risk prediction using DMD-based graph neural networks.

Wen Xu1, Changjun Wang2, Yuhang Chu3

  • 1School of Traffic Management, People's Public Security University of China, Beijing 100038, China; Intelligent Policing Key Laboratory of Sichuan Province, Luzhou 646000, China; Sichuan Police College, Luzhou 646000, China.

Accident; Analysis and Prevention
|March 30, 2026
PubMed
Summary

Related Concept Videos

You might also read

Related Articles

Articles linked to this work by shared authors, journal, and citation graph.

Sort by
Same author

Development of cinnamon essential oil microcapsules with enhanced antimicrobial activity via interfacial crosslinking of OSA-modified starch nanocrystal stabilized Pickering emulsions.

International journal of biological macromolecules·2026
Same author

Generation and evaluation of an EcoHIV mouse model to study HIV-1 integration and expression.

Virology·2026
Same author

Modulating burn wound immunity and vasculature with a photo-crosslinkable glycyrrhizic acid-ginsenoside hydrogel.

Biomedical materials (Bristol, England)·2026
Same author

The epidemic and socioeconomic factors shock: a long-term dynamic analysis of equity in infectious disease healthcare in China.

International journal for equity in health·2026
Same author

Genomic insights into the tmexCD-toprJ: plasmid-mediated evolution, dissemination and diversity in bacterial populations.

The Journal of antimicrobial chemotherapy·2026
Same author

Urban high-risk scenarios for automated vehicle safety testing: A generation and generalization method based on accident data.

Accident; analysis and prevention·2026

This study introduces a hybrid model combining Hankel-based Dynamic Mode Decomposition (Hankel-DMD) and spatiotemporal graph neural networks (STGNN) for accurate urban traffic crash prediction. The novel approach enhances proactive traffic safety management by providing interpretable, neighborhood-level forecasts.

Area of Science:

  • Urban planning and traffic safety
  • Data science and machine learning
  • Spatiotemporal modeling

Background:

  • Accurate spatiotemporal prediction of urban traffic crashes is crucial for proactive safety management.
  • Existing models often struggle to capture both temporal dynamics and spatial dependencies in an interpretable manner.
  • Neighborhood-level crash prediction requires sophisticated methods to handle complex urban networks.

Purpose of the Study:

  • To develop a hybrid framework integrating Hankel-DMD with STGNN for short-term, neighborhood-level traffic crash count prediction.
  • To improve the interpretability and accuracy of spatiotemporal traffic crash forecasting models.
  • To provide a tool for targeted urban safety interventions.

Main Methods:

  • A hybrid framework combining Hankel-based Dynamic Mode Decomposition (Hankel-DMD) and spatiotemporal graph neural networks (STGNN).
Keywords:
Crashrisk predictionGraph neural networksHankel dynamic mode decompositionInterpretable deep learningSpatiotemporal modelingUrban traffic safety

Related Experiment Videos

  • Application of Hankel-DMD to a neighborhood-day crash matrix to extract dominant spatiotemporal modes.
  • Utilizing an STGNN on a graph representing neighborhood relationships to learn nonlinear residuals and spatial dependencies.
  • Multi-step prediction evaluation across various forecast horizons (1-7 days).

    • Main Results:

    • The hybrid model significantly outperformed statistical, tree-based, pure Hankel-DMD, and baseline STGNN models in multi-step crash count prediction.
    • Achieved 17-30% lower mean absolute error and 13-24% lower root mean squared error compared to the best deep learning benchmark.
    • Demonstrated consistent performance gains across high-, medium-, and low-risk neighborhood groups.
    • Hankel-DMD modes revealed stable urban structures in 2019/2021 and deviations in 2020 linked to mobility changes.

    Conclusions:

    • Dynamics-informed graph learning provides accurate and interpretable crash risk forecasts at the neighborhood scale.
    • The hybrid Hankel-DMD-STGNN model offers a powerful approach for proactive urban traffic safety management.
    • The findings support the development of targeted safety interventions based on reliable spatiotemporal crash predictions.