Scholarlink

<p>The effectiveness of clustering trajectories is heavily dependent on the selected (dis)similarity function. While previous studies have primarily focused on distance, incorporating additional geometric features—such as shape, sinuosity, complexity, orientation, and buffer overlap area—can significantly enhance clustering quality. In this study, we present an approach to determine the optimal weight for each geometric feature to maximize clustering quality. We evaluate the performance of grasshopper, grey wolf, and particle swarm optimization algorithms using real-world datasets that include pedestrian, car, vessel, and flight tracks. Our findings demonstrate that the proposed approach improves clustering quality by 20.11%, 20.08%, and 15.65% compared to relying solely on the distance feature. Among the algorithms, the grasshopper optimization algorithm achieved the highest clustering quality, outperforming both the grey wolf and particle swarm algorithms. Notably, distance, orientation, buffer overlap area, and complexity emerged as the most influential features in trajectory clustering.</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

Enhancing clustering of trajectories through optimization of geometric features

  • 期刊论文

  • 发表日期：

    2025年2月19日

查看全文
  • Pooya Shivanasab,
  • Rahim Ali Abbaspour,
  • Alireza Chehreghan

摘要

The effectiveness of clustering trajectories is heavily dependent on the selected (dis)similarity function. While previous studies have primarily focused on distance, incorporating additional geometric features—such as shape, sinuosity, complexity, orientation, and buffer overlap area—can significantly enhance clustering quality. In this study, we present an approach to determine the optimal weight for each geometric feature to maximize clustering quality. We evaluate the performance of grasshopper, grey wolf, and particle swarm optimization algorithms using real-world datasets that include pedestrian, car, vessel, and flight tracks. Our findings demonstrate that the proposed approach improves clustering quality by 20.11%, 20.08%, and 15.65% compared to relying solely on the distance feature. Among the algorithms, the grasshopper optimization algorithm achieved the highest clustering quality, outperforming both the grey wolf and particle swarm algorithms. Notably, distance, orientation, buffer overlap area, and complexity emerged as the most influential features in trajectory clustering.