Multidimensional Data Mining of Traffic Anomalies on Large-Scale Road Networks

The identification and characterization of traffic anomalies on massive road networks is a vital component of traffic monitoring and control. Anomaly identification can be used to reduce congestion, increase safety, and provide transportation engineers with better information for traffic forecasting and road network design. However, because of the size, complexity, and dynamics of transportation networks, automating such a process is challenging. A multidimensional mining framework is proposed; it can be used to identify a concise set of anomalies from massive traffic monitoring data and then overlay, contrast, and explore such anomalies in multidimensional space. The framework is based on the development of two novel methods: (1) efficient anomaly mining stemming from the discovery of the atypical fragment (a compact representation of a set of abnormal traffic patterns happening across a sequence of connected road segments, possibly spanning multiple roads, and occurring at overlapping time intervals) and (2) a multidimensional anomaly overlay model that enables the clustering of multiple atypical fragments according to different criteria (e.g., severity, topology, or spatiotemporal characteristics). The atypical fragment provides a concise, global view of the traffic anomaly situation, whereas the framework for anomaly overlay provides the power of online analytical processing to facilitate the discovery of patterns associated with different anomaly types and the navigation of anomalies at multilevel abstraction.