计算机工程与应用 ›› 2022, Vol. 58 ›› Issue (1): 175-181.DOI: 10.3778/j.issn.1002-8331.2007-0298

• 模式识别与人工智能 • 上一篇    下一篇

基于占空比的聚类算法评价指标研究

张欣环,刘宏杰,吴金洪,施俊庆,毛程远,孟国连   

  1. 1.浙江师范大学 道路与交通工程研究中心,浙江 金华 321004
    2.西安交通大学 电子信息工程学院,西安 710049
  • 出版日期:2022-01-01 发布日期:2022-01-06

Research on Evaluation Function of Clustering Algorithm Based on Duty Cycle#br#

ZHANG Xinhuan, LIU Hongjie, WU Jinhong, SHI Junqing, MAO Chengyuan, MENG Guolian   

  1. 1.Road and Traffic Engineering Institute, Zhejiang Normal University, Jinhua, Zhejiang 321004, China
    2.The School of Electronic and Information Engineering, Xi’an Jiao Tong University, Xi’an 710049, China
  • Online:2022-01-01 Published:2022-01-06

摘要: 基于密度的聚类算法(DBSCAN)是最有效的轨迹数据挖掘方法之一,但基于密度的聚类算法往往受到输入参数选择的限制。在轨迹数据挖掘中,聚类结果不仅受到类内距离和类间距离的影响,还受到聚类中坐标点个数的影响。因此,提出了一种新的基于内外占空比的集群有效性指标来平衡这三个因素,该指标可以自动选择密度聚类的输入参数,并在不同的数据集上形成有效的聚类,优化后的聚类方法可应用于出行者行为轨迹的深度分析和挖掘。实验结果证明,与传统的有效性指标相比,提出的基于占空比的评价指标能够优化输入参数,获得较好的出行者位置信息聚类结果。

关键词: DBSCAN算法, 有效性指数, 密度聚类, 轨迹聚类

Abstract: Density-based clustering (DBSCAN) is one of the most effective methods for trajectory data mining, but density-based clustering algorithms are often limited by the choice of input parameters. In the trajectory data mining, clustering results are not only affected by the within-class distance and between-class distance, but also by the number of coordinate points in the cluster. Therefore, this paper proposes a novel cluster validity index based on the internal and external duty cycle to balance the three parameters. In this way, the parameters of density clustering can be automatically selected, and effective clustering can be formed on different datasets. Then the clustering method is applied to the depth analysis and mining of travelers’ behavior trajectories. The experiments prove that compared with the traditional validity index, the evaluation function proposed can optimize input parameters and get better clustering results of user location information.

Key words: DBSCAN algorithm, validity index, density-based clustering, trajectory clustering