A binary logit-based incident detection model for urban traffic networks

Yaser E. Hawas, Faisal Ahmed

Research output: Contribution to journalArticlepeer-review

4 Citations (Scopus)


Incident detection systems for the urban traffic network are still lacking efficient models for better performance or efficiency to detect incidents of relatively short durations, as well as stability of performance or transferability and applicability to generalized network conditions. This paper presents a new urban incident detection system based on a binary-logit regression model. The distinctive feature of the presented model is its ability to detect incidents of relatively short durations. Two sets of data were generated for various scenarios of incidents on different lanes using microscopic simulation. Each data-set is further divided into two sub-sets: one for calibration and another for validation purposes. The proposed logit model incorporates various traffic link flows, signal green phases and cycle times, and link lengths as the input parameters along with the extracted detector readings. The logit model, coupled with some predefined threshold value, estimates the incident status probability at each cycle time (or analysis time step). Detection rates of 80.7 and 74.8% were obtained from the two data-sets. The mean response time to detect an incident and the values of false alarm rates were found to be quite satisfactory. An extensive sensitivity analysis is conducted to study the impact of all of the input parameters (or the independent variables of the model).

Original languageEnglish
Pages (from-to)49-62
Number of pages14
JournalTransportation Letters
Issue number1
Publication statusPublished - Jan 1 2017


  • Binary logit model
  • Detection rate
  • False alarm rate
  • Lane detectors
  • Mean time to detect
  • Urban incident detection

ASJC Scopus subject areas

  • Transportation


Dive into the research topics of 'A binary logit-based incident detection model for urban traffic networks'. Together they form a unique fingerprint.

Cite this