Scholarlink

<p>While the link between driver behavior and accident causes is well-established, quantifying the complex, mediating pathways of human factors within an integrated, multidimensional framework remains a key challenge. This study aims to provide an integrated framework for understanding these interactions. Using data from the Strategic Highway Research Program Phase II Naturalistic Driving Study (SHRP2 NDS), we first constructed a multidimensional accident causation dataset. Subsequently, we developed a causation-accident association model using structural equation modeling (SEM) to quantify the relationships between multiple risk factors, with particular attention to the mediating effects of human factors. Building upon these established relationships, we developed an innovative accident risk prediction technology that integrates a voting-based ensemble method with a genetic algorithm-optimized neural network (NN). The results reveal that driver performance significantly impacts accident probability through both direct effects and complex mediating pathways involving demographic characteristics, road conditions, and environmental factors. Ultimately, the proposed prediction model achieves 96% accuracy and an F1 score of 87%, outperforming traditional approaches by 4–13% in terms of F1-score. This study not only advances our understanding of accident causation mechanisms but also provides an effective technological framework for real-time risk assessment and accident prevention.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Risk prediction of road traffic accidents: a human-centered multi-factor analysis framework

  • 期刊论文

  • 发表日期：

    2025年10月13日

查看全文
  • Jing Huang,
  • Jinleng Zhu,
  • Zilin Chen,
  • Lin Hu

摘要

While the link between driver behavior and accident causes is well-established, quantifying the complex, mediating pathways of human factors within an integrated, multidimensional framework remains a key challenge. This study aims to provide an integrated framework for understanding these interactions. Using data from the Strategic Highway Research Program Phase II Naturalistic Driving Study (SHRP2 NDS), we first constructed a multidimensional accident causation dataset. Subsequently, we developed a causation-accident association model using structural equation modeling (SEM) to quantify the relationships between multiple risk factors, with particular attention to the mediating effects of human factors. Building upon these established relationships, we developed an innovative accident risk prediction technology that integrates a voting-based ensemble method with a genetic algorithm-optimized neural network (NN). The results reveal that driver performance significantly impacts accident probability through both direct effects and complex mediating pathways involving demographic characteristics, road conditions, and environmental factors. Ultimately, the proposed prediction model achieves 96% accuracy and an F1 score of 87%, outperforming traditional approaches by 4–13% in terms of F1-score. This study not only advances our understanding of accident causation mechanisms but also provides an effective technological framework for real-time risk assessment and accident prevention.