文章基本信息

标题：Outlier Detection and Comparison of Origin-Destination Flows Using Data Depth
本地全文：下载
作者：Myeong-Hun Jeong ; Junjun Yin ; Shaowen Wang 等
期刊名称：LIPIcs : Leibniz International Proceedings in Informatics
电子版ISSN：1868-8969
出版年度：2018
卷号：114
页码：1-14
DOI：10.4230/LIPIcs.GISCIENCE.2018.6
出版社：Schloss Dagstuhl -- Leibniz-Zentrum fuer Informatik
摘要：Advances in location-aware technology have resulted in massive trajectory data. Origin-destination (OD) trajectories provide rich information on urban flow and transport demand. This study describes a new method for detecting OD flows outliers and conducting hypothesis testing between two OD flow datasets in terms of the variations of spatial extent, that is, spread. The proposed method is based on data depth, which measures the centrality and outlyingness of a point with respect to a given dataset in R^d. Based on the center-outward ordering property, the proposed method analyzes the underlying characteristics of OD flows, such as location, outlyingness, and spread. The ability of the method to detect OD anomalies is compared with that of the Mahalanobis distance approach, and an F-test is used to verify the difference in scale. Empirical evaluation has demonstrated that our method effectively identifies OD flows outliers in an interactive way. Furthermore, the method can provide new perspectives such as spatial extent by considering the overall structure of data when comparing two different OD flows in terms of scale.
关键词：Movement Analysis; Trajectory Data Mining; Data Depth; Outlier Detection