cicyt UNIZAR
Full-text links:

Download:

Current browse context:

physics.soc-ph

Change to browse by:

References & Citations

Bookmark

(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Physics > Physics and Society

Title: Anomaly Detection in Road Networks using Sliding-Window Tensor Factorization

Abstract: Anomaly detection on road networks can be used to sever for emergency response and is of great importance to traffic management. However, none of the existing approaches can deal with the diversity of anomaly types. In this paper, we propose a novel framework to detect multiple types of anomalies. The framework incorporates real-time and historical traffic into a tensor model and acquires spatial and different scale temporal pattern of traffic in unified model using tensor factorization. Furthermore, we propose a sliding window tensor factorization to improve the computational efficiency. Basing on this, we can identify different anomaly types through measuring the deviation from different spatial and temporal pattern. Then, to promote a deeper understanding of the detected anomalies, we use an optimization method to discover the path-level anomalies. The core idea is that the anomalous path inference is formulated as L1 inverse problem by considering the sparsity of anomalies and flow on paths simultaneously. We conduct synthetic experiments and real case studies based on a real-world dataset of taxi trajectories. Experiments verify that the proposed framework outperforms all baseline methods on efficiency and effectiveness, and the framework can provide a better understanding for anomalous events.
Subjects: Physics and Society (physics.soc-ph); Social and Information Networks (cs.SI)
Cite as: arXiv:1803.04534 [physics.soc-ph]
  (or arXiv:1803.04534v1 [physics.soc-ph] for this version)

Submission history

From: Ming Xu [view email]
[v1] Wed, 7 Mar 2018 10:07:27 GMT (1537kb)