Synapse
⌘+K
Synapse
PulseExploreClubsResearchersJournals
Instagram
HomeClubsExplore
May 16, 2026Accident Analysis & PreventionOpen Access

LLM-enhanced causal graph learning for real-time crash risk prediction

View Full Paper
Ask AI
Bookmark
Share

Authors

CLChenguang LiCentral South UniversityHHHelai HuangCentral South UniversityHZHanchu ZhouCentral South University

Discussion

Loading...

Member takes

Implication

Randomized trial demonstrates enhanced predictive accuracy in collision risk prediction, suggesting real-time traffic management applications.

Key Points

  • This study aims to improve traffic crash risk prediction by integrating causal learning with spatio-temporal models.
  • Developed a closed-loop framework combining causal discovery, semantic enhancement, and spatio-temporal prediction.
  • Utilized transfer entropy to extract causal relationships and refined causal graphs using a GPT-2-based model with LoRA.
  • Merged enhanced graphs with temporal features using Graph Attention Networks and Long Short-Term Memory networks for prediction.
  • Achieved an accuracy of 0.956, F1-score of 0.872, and AUC of 0.985, outperforming traditional models.
  • GPT-2-enhanced causal graphs contributed significantly to predictive accuracy, confirmed through ablation studies.

Cite This Study

Li et al. (2026) studied this question.

synapsesocial.com/papers/6a0808ffa487c87a6a40b1cdhttps://doi.org/10.1016/j.aap.2026.108579
View Full Paper
Ask AI
Bookmark
Share

Also Consider

Synapse has enriched 4 closely related papers on similar clinical questions. Consider them for comparative context:

  1. 1A causal discovery approach to study key mixed traffic‐related factors and age of highway affecting raveling2024 · 2 citations
  2. 2Statistical estimation of fatal and serious injuries saved by iRAP protocols in 74 countries2024 · 13 citations
  3. 3Incorporating multimodal context information into traffic speed forecasting through graph deep learning2023 · 46 citations
  4. 4Intelligent Traffic Flow Prediction Using Optimized GRU Model2021 · 104 citations