Incident Duration Time Prediction Using Supervised Topic Modeling Method

Research output: Chapter in Book/Report/Conference proceedingChapter


Accurate prediction of the duration of traffic incidents is one of the most prominent prerequisites for effective implementation of proactive traffic incident management strategies. This paper presents a novel method for immediate prediction of traffic incident duration using an emerging supervised topic modeling. The proposed method employs natural language processing techniques for semantic text analysis of the text-based incident traffic incident dataset. The model applies the labeled latent Dirichlet allocation approach, and it is trained using 1,466 incident records collected by the Korea Expressway Corporation from 2016 to 2019. For training purposes, the proposed method divides the incidents into two groups based on the incident duration: incidents shorter than 2 h and incidents lasting 2 h or longer, following the incident management guidelines of the Federal Highway Administration Manual on Uniform Traffic Control Devices for Streets and Highways (2009). The model is tested with randomly selected incident records that were not used for the model training. The results demonstrate overall prediction accuracies of approximately 74% for incidents lasting up to 2 h, and 82% for incidents lasting 2 h or longer.

Original languageEnglish (US)
Title of host publicationTransportation Research Record
PublisherSAGE Publications Ltd
Number of pages13
StatePublished - Feb 2023

All Science Journal Classification (ASJC) codes

  • Civil and Structural Engineering
  • Mechanical Engineering


  • artificial intelligence
  • artificial intelligence and advanced computing applications
  • automated reasoning
  • data and data science
  • deep learning
  • freeway operations
  • incident management
  • machine learning (artificial intelligence)
  • motorcycles and mopeds
  • operations
  • safety
  • safety data
  • supervised learning


Dive into the research topics of 'Incident Duration Time Prediction Using Supervised Topic Modeling Method'. Together they form a unique fingerprint.

Cite this