工程科技文献索引服务平台 | abstract.cmanuf.com

Anomaly Detection on ITS Data via View Association

刊名：	SIGKDD explorations
作者：	Junaidillah Fadlil（National Taiwan University of Science and Technology） Hsing-Kuo Pao（National Taiwan University of Science and Technology） Yuh-Jye Lee（National Taiwan University of Science and Technology）
刊号：	739B0151
出版年：	2013
年卷期：	2013, no.CD/ROM
页码：	470-478
总页数：	9
分类号：	TP274.2
关键词：	Anomaly Detection；Intelligent Transportation System (ITS)；Association；Trajectory；Manifold learning
参考中译：
语种：	eng
文摘：	We focus on detecting anomalous events in transportation systems. In transportation systems, other than normal road situation, anomalous events happen once in a while such as traffic accidents, ambulance car passing, harsh weather conditions, etc. Identifying the anomalous traffic events is essential because the events can lead to critical conditions where immediate investigation and recovery may be necessary. We propose an anomaly detection method for transportation systems where we create a police report automatically after detecting anomalies. Unlike the traditional police report, in this case, some quantitative analysis shall be done as well to provide experts with an advanced, precise and professional description of the anomalous event. For in- stance, we can provide the moment, the location as well as how severe the accident occurs in the upstream and down- stream routes. We present an anomaly detection approach based on view association given multiple feature views on the transportation data if the views are more or less independent from each other. For each single view, anomalies are detected based on a manifold learning and hierarchical clustering procedures and anomalies from different views are associated and detected as anomalies with high confidence. We study two well-known ITS datasets which include the data from Mobile Century project and the PeMS dataset, and we evaluate the proposed method by comparing the automatically generated report and real report from police during the related period.