What is it about?
Applying Mixed Logit Model (MXL) and Support Vector Machine (SVM), this study investigates the factors that impact the severity levels of work zone crashes.
Featured Image
Photo by Jamie Street on Unsplash
Why is it important?
Work zones have been a high-priority issue due to their impacts on traffic safety. A better understanding of work zone crashes can help to identify the contributing factors and countermeasures in order to enhance roadway safety. Florida is among the top three states across the nation with the highest fatality rates. Miami-Dade County had the highest crash rates in the state of Florida, with a total of 64,070 fatality and injury crashes in 2016. The number of work zones has increased due to the growth of highway renovations and construction projects in the state of Florida. The number of crashes associated with work zones increased from 1,153 in 2013 to 1,315 in 2017. Thus, safety should be an important consideration by decision makers; as they plan, design, and operate the work zone.
Perspectives
The contribution of this study lies in two major tasks. First, work zone crash injury severity was investigated through a mixed logit modeling approach as a parametric model. Second, by applying to the same dataset, support vector machine models were developed in order to predict crash injury severity. In addition, enhancement of the SVM model performance was investigated not only by examining different kernel functions, but through applying three different optimization algorithms. Comparing the prediction accuracy of the two proposed models revealed that the improved SVM model (i.e., HS-SVM) produced better prediction results than the MXL model, which was 83.5%, compared to 67.2%.
Seyedmirsajad Mokhtarimousavi
Florida International University
Read the Original
This page is a summary of: Improved Support Vector Machine Models for Work Zone Crash Injury Severity Prediction and Analysis, Transportation Research Record Journal of the Transportation Research Board, June 2019, SAGE Publications,
DOI: 10.1177/0361198119845899.
You can read the full text:
Contributors
The following have contributed to this page
