Fast fuzzy trajectory clustering strategy based on data summarization and rough approximation

详细信息    查看全文

• 作者：Chunchun Hu ; Nianxue Luo ; Qiansheng Zhao
• 关键词：Fuzzy clustering ; Trajectory dataset ; Rough set ; Data summarization ; The membership degree
• 刊名：Cluster Computing
• 出版年：2016
• 出版时间：September 2016
• 年：2016
• 卷：19
• 期：3
• 页码：1411-1420
• 全文大小：1,529 KB
• 刊物类别：Computer Science
• 刊物主题：Processor Architectures
  Operating Systems
  Computer Communication Networks
  
• 出版者：Springer Netherlands
• ISSN：1573-7543
• 卷排序：19

文摘

More technologies available for collecting large dataset of trajectory of moving objects make it more essential to perform clustering over these trajectory data. Trajectory clustering presents more complication than traditional approaches due to the nature of trajectories which is temporal, massive and related to the location. Meanwhile, the uncertainty in trajectory clustering also appears when we determine which cluster each trajectory belongs to. However, the computing performance of many clustering algorithms sharply declines as data size increases. In this paper, we study the fuzzy clustering approach for extracting potential spatial patterns by introducing rough set and fuzzy set theory. First, we propose the fast similarity measure method by employing the rough approximation distances between trajectories. Especially for the long trajectory sequences, the computing time would be reduced greatly. We also introduce a summarization technique to reduce the number of distance computations required in similarity measure. Second, an appropriate function of the membership degree is redefined for clustering quality and performance purpose. Third, we modify the fuzzy C-means algorithm by embodying a new similarity measure and the membership degree function. The experimental results conducted on two real datasets of trajectories show the effectiveness of our methods by evaluating clustering validity and computing performance for large datasets. The computing performance of the proposed fuzzy clustering is obviously improved as the dataset size of trajectories increases.