Scholarlink

<p>Road traffic accidents represent a significant safety concern worldwide, ranking among the leading causes of fatalities and a major risk to human life and property. Addressing this challenge requires a comprehensive understanding of the factors contributing to accidents and their consequences. This study uses advanced data mining and analysis techniques to investigate data from traffic accidents, focusing on temporal, spatial, environmental, and road-related factors. The research aims to uncover hidden patterns within the data and provide a scientific basis for proactive accident prevention and efficient post-accident response. The study begins with a comprehensive preprocessing workflow, including data reduction, transformation, and cleaning, ensuring the validity and reliability of the data. Temporal, spatial, and environmental characteristics, such as time, location, weather, temperature, and lighting, are analyzed and visualized to understand traffic accident distribution patterns comprehensively. An association analysis is then performed using the <i>K</i>-means clustering algorithm to discretize continuous accident variables, followed by applying the Apriori algorithm to uncover hidden relationships among accident causes. An accident severity-weighted quantization method enhances the algorithm, effectively identifying significant association rules related to severe accidents. Based on these findings, targeted accident prevention strategies are proposed. Finally, the study evaluates multiple deep learning and machine learning algorithms for accident duration prediction, identifying optimized support vector regression via Bayesian and particle swarm optimization (BPSO-SVR) as the best-performing model. The optimized BPSO-SVR model combines Bayesian optimization (BO) and particle swarm optimization (PSO) to fine-tune model hyperparameters, significantly improving prediction accuracy. An interpretability analysis identifies key factors influencing accident duration, including severity, crossing characteristics, sunrise/sunset conditions, and weather. These findings contribute to advancing road traffic safety research, offering valuable insights for early-warning systems and efficient post-accident interventions.</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

Data-driven analysis and prediction of traffic accident dynamics using spatiotemporal modeling and optimized machine learning techniques

  • 期刊论文

  • 发表日期：

    2026年1月12日

查看全文
  • Doha Turab,
  • Sharjeel Muzaffar,
  • Lubna Nadeem,
  • Jianqiang Li,
  • Yu Wang,
  • Tariq Mahmood,
  • Tanzila Saba

摘要

Road traffic accidents represent a significant safety concern worldwide, ranking among the leading causes of fatalities and a major risk to human life and property. Addressing this challenge requires a comprehensive understanding of the factors contributing to accidents and their consequences. This study uses advanced data mining and analysis techniques to investigate data from traffic accidents, focusing on temporal, spatial, environmental, and road-related factors. The research aims to uncover hidden patterns within the data and provide a scientific basis for proactive accident prevention and efficient post-accident response. The study begins with a comprehensive preprocessing workflow, including data reduction, transformation, and cleaning, ensuring the validity and reliability of the data. Temporal, spatial, and environmental characteristics, such as time, location, weather, temperature, and lighting, are analyzed and visualized to understand traffic accident distribution patterns comprehensively. An association analysis is then performed using the K-means clustering algorithm to discretize continuous accident variables, followed by applying the Apriori algorithm to uncover hidden relationships among accident causes. An accident severity-weighted quantization method enhances the algorithm, effectively identifying significant association rules related to severe accidents. Based on these findings, targeted accident prevention strategies are proposed. Finally, the study evaluates multiple deep learning and machine learning algorithms for accident duration prediction, identifying optimized support vector regression via Bayesian and particle swarm optimization (BPSO-SVR) as the best-performing model. The optimized BPSO-SVR model combines Bayesian optimization (BO) and particle swarm optimization (PSO) to fine-tune model hyperparameters, significantly improving prediction accuracy. An interpretability analysis identifies key factors influencing accident duration, including severity, crossing characteristics, sunrise/sunset conditions, and weather. These findings contribute to advancing road traffic safety research, offering valuable insights for early-warning systems and efficient post-accident interventions.