Predicting Incident Duration based on Spatiotemporal Heterogeneous Pattern Recognition

Traffic incidents are responsible for a significant number of non-recurrent congestions. Accurate prediction of incident duration can help mitigate congestions, reduce incident-induced delays, as well as improve the travel time reliability. Similar to roadway traffic flow status, incident duration can have specific features associated with location and time, which is often referred to as the spatiotemporal heterogeneity. This study analyzes the incident duration data on I-5 in Washington State. A two-step methodology framework, including a pattern recognition procedure and survival analysis model, has been developed to predict incident durations. For each incident of interest, the pattern recognition algorithm, namely, the k-Nearest Neighboring, is firstly performed to identify a group of spatiotemporally “nearest” incidents. These selected incidents have then been introduced in the hazard-based survival analysis to fit a duration prediction model. Results show that spatiotemporal pattern recognition helps improve the model performance, in the sense that duration estimates based on the k-NN selected incidents are more accurate than the general model based on all historical incidents. The results are consistent for different distribution assumptions (log-logistic, log-normal, and Weibull distribution) of the incident duration data. Pattern recognition also provides a computationally efficient solution, as a smaller number of data are involved in duration prediction modeling. For the studied dataset, accurate duration prediction can be achieved by modeling the 500 most relevant incident records, which approximately equals to 1/100 of the entire dataset. The proposed methodology framework can work dynamically with a large-scale incident database.

• Supplemental Notes:
  • This paper was sponsored by TRB committee ANB20 Standing Committee on Safety Data, Analysis and Evaluation.
• Corporate Authors:

  Transportation Research Board

  500 Fifth Street, NW
  Washington, DC  United States  20001
• Authors:
  • Zhu, Wenbo
  • Zeng, Ziqiang
  • 0000-0003-1782-2804
  • Wang, Yinhai
  • Pu, Ziyuan
• Conference:
  • Transportation Research Board 96th Annual Meeting
  • Location: Washington DC, United States
  • Date: 2017-1-8 to 2017-1-12
• Date: 2017

Language

• English

Media Info

• Media Type: Digital/other
• Features: Figures; References; Tables;
• Pagination: 16p
• Monograph Title: TRB 96th Annual Meeting Compendium of Papers

Subject/Index Terms

• TRT Terms: Mathematical models; Pattern recognition systems; Time duration; Traffic congestion; Traffic incidents; Travel time
• Geographic Terms: Washington (State)
• Subject Areas: Highways; Safety and Human Factors;

Filing Info

• Accession Number: 01628218
• Record Type: Publication
• Report/Paper Numbers: 17-06787
• Files: TRIS, TRB, ATRI
• Created Date: Mar 7 2017 10:25AM