Improve Model Robustness of Traffic Crash Risk Evaluation Via Adversarial Mix-Up Under Traffic Flow Fundamental Diagram

SSRN Electronic Journal(2023)

引用 0|浏览8
暂无评分
摘要
•Proposed a traffic flow adversarial example (TF-AE) generation method via mix-up under traffic flow fundamental diagram.•With the developed TF-AEs to evaluate model robustness, model accuracy decreased by 8.0% and sensitivity dropped by 18.0%.•Developed a coverage-oriented adversarial training to improve robustness, avoiding 76% accuracy, 98% sensitivity drops.•The crash risk evaluation model with adversarial training had more stable outputs to real-world traffic dynamic fluctuations.
更多
查看译文
AI 理解论文
溯源树
样例
生成溯源树,研究论文发展脉络