Scholarlink

<p>The timing of traffic lights has become one of the most daunting tasks in efficiently controlling the urban traffic system junctions. Complex and dynamic traffic is the root of poorly timed signals that lead to delays, congestion, and inefficiency. The authors propose a new hybrid framework incorporating XGBoost, ST-GNN, and PSO to handle such challenges. XGBoost is used to accurately predict traffic flow and identify the most critical features influencing signal timing. ST-GNNs extend these predictions by modeling spatial–temporal dependencies in traffic patterns, including interaction among intersections over time. Finally, PSO finds a dynamic optimization of signal timings to minimize delays and optimize throughput based on real-time predictions. These results confirm that the proposed framework outperforms the classic methods, offering an average network delay reduction of 29.6%, with over 20% more throughput. This can be indicated as an integrated approach that could lead to scalable, adaptive, data-driven management of traffic lights, opening a new paradigm for efficient and sustainable urban mobility solutions.</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

Machine learning-based prediction of traffic signal timing for optimized intersection management

  • 期刊论文

  • 发表日期：

    2025年8月13日

查看全文
  • Amani Abdallah Assolie,
  • Ibrahim Khliefat,
  • Ali Alqatawna,
  • Eslam Mohammad AL-kharabsheh,
  • Salam AL Kasassbeh

摘要

The timing of traffic lights has become one of the most daunting tasks in efficiently controlling the urban traffic system junctions. Complex and dynamic traffic is the root of poorly timed signals that lead to delays, congestion, and inefficiency. The authors propose a new hybrid framework incorporating XGBoost, ST-GNN, and PSO to handle such challenges. XGBoost is used to accurately predict traffic flow and identify the most critical features influencing signal timing. ST-GNNs extend these predictions by modeling spatial–temporal dependencies in traffic patterns, including interaction among intersections over time. Finally, PSO finds a dynamic optimization of signal timings to minimize delays and optimize throughput based on real-time predictions. These results confirm that the proposed framework outperforms the classic methods, offering an average network delay reduction of 29.6%, with over 20% more throughput. This can be indicated as an integrated approach that could lead to scalable, adaptive, data-driven management of traffic lights, opening a new paradigm for efficient and sustainable urban mobility solutions.