Modeling Crash Severity and Collision Types Using Machine Learning

Traffic crashes are an important source of non-recurrent congestion causing delays to travelers in a transportation network. In the year 2017, Texas recorded 1.38 death per 100 million miles traveled. The United States’ annual average cost of road crashes is estimated to be around $230.6 billion or around $820 per person. Typically, crash analysis is complex due the presence of human behavior element which is difficult to predict and model. In the context of traffic safety, understanding the factors affecting crash occurrence, injury severity and collision type and their underlying relationships help us in predicting future crashes, its severity and collision type under given circumstances. The analysis of traffic safety often involves categorization or classification problems. Literature in the domain of traffic safety presents two classification systems. First, categorizes crash based on the crash severity as: no injury, possible injury, non-incapacitating injury, incapacitating injury and fatal injury crashes. Second classify crash based on collision type as: rear end, side swipe, angular, opposite direction and single motor vehicle crashes. Past studies have attempted to model the collision type and crash severity type considering a number of explanatory variables affecting crash occurrence, injury severity and collision type. Past studies typically approach to model the crashes for severity types and collision types separately. However, the severity types and collision types may be correlated and it is difficult to factor-in these correlation if the severity types and collision types are modeled separately. Although past studies recognizes these correlations and suggest the need to factor-in these correlation in modeling crashes, there has been limited attempt to incorporate this correlation in modeling process. Modeling them separately necessitate more complex model structure to account for cross-model-correlations. This research is motivated to bridge this gap in literature and aims to model collision type and crash severity type simultaneously recognizing the fact that they may be correlated. The insights from these safety analysis and model outputs will help in identifying critical locations/links in a transportation network. The information about critical links can be used for optimal positioning of troopers, and in prioritizing the location for frequent surveillance by traffic management centers. In particular, the model can be used for predicting the probable locations of crashes and severity types. This will allow troopers to position themselves in strategic locations. As troopers will be nearby to incident location they are more likely to reach the incident spot in smaller time and help stabilize the victim in their golden hour. This will also help in clearing the traffic in shorter time thereby saving fuel and reducing air pollution due to congestion built by incident which otherwise may be for prolonged time. It can also help in better planning for incident management and in optimal allocation of resources/funds.

• Record URL:
  https://transet.lsu.edu/research-in-progress/
• Supplemental Notes:
  • 20SAUTSA36

Language

• English

Project

• Status: Completed
• Funding: $ 80010
• Contract Numbers:

  69A3551747106

• Sponsor Organizations:

  Office of the Assistant Secretary for Research and Technology

  University Transportation Centers Program
  Department of Transportation
  Washington, DC  United States  20590
• Managing Organizations:

  Transportation Consortium of South-Central States (Tran-SET)

  Louisiana State University
  Baton Rouge, LA  United States  70803
• Project Managers:

  Mousa, Momen

• Performing Organizations:

  University of Texas at San Antonio

  One UTSA Circle
  San Antonio, TX  United States  78249
• Principal Investigators:

  Kumar, Amit

• Start Date: 20200801
• Expected Completion Date: 20220401
• Actual Completion Date: 0
• USDOT Program: University Transportation Centers Program

Subject/Index Terms

• TRT Terms: Crash analysis; Crash severity; High risk locations; Machine learning; Traffic safety
• Subject Areas: Highways; Safety and Human Factors;

Filing Info

• Accession Number: 01757542
• Record Type: Research project
• Source Agency: Transportation Consortium of South-Central States (Tran-SET)
• Contract Numbers: 69A3551747106
• Files: UTC, RIP
• Created Date: Nov 11 2020 9:26AM