An Implementation Framework for Random Parameter Models for Crash Frequency Prediction and Safety Investment Prioritization on the Washington State DOT Highway Network

The Washington State Department of Transportation (WSDOT) highway network is comprised of multiple functional classes, ranging from two-lane rural to two-lane urban, interstate, and other classes. This study focused on developing a framework for implementing random parameter models of annual crash frequency for these roadway classes. The implementation framework includes statistical models developed for the two-year period 2014-2015 at the 1-mile scale, and a web-based interface that embeds the 1-mile models in a user-friendly manner. The objective of the web-based interface is to provide to decision makers a facility to evaluate the impact of changes in geometric attributes on predicted crash counts for any segment on the WSDOT highway network. The total number of segments for each functional class is defined by 1-mile segment counts. A 1-mile segmentation scale was used to ensure project-level consistency with decisions arising from a random parameter statistical model. In the literature, a variety of segmentations have been analyzed—from homogeneous segments (which can be as small as 0.01 miles) to segmentation that includes homogeneous segments with respect to curvature, to segmentation that is fixed-length in nature (for example, 1 mile). The 1-mile segmentation provides the facility to analyze the impact of corridor-level safety investments. This is done due to the fact that heterogeneity in the corridors is addressed at a higher scale than the typical homogeneous segment scale. The 1-mile scale also enables compatibility with network screening methods that are widely used in the application of the Highway Safety Manual (HSM). A demonstration of the implementation framework is presented in the form of a dashboard system. The demonstration system provides the opportunity to explore enhancements to maximize its utility. Recommendations include soliciting input to leverage the rich information in safety models when planning infrastructure improvements; examining other modelling approaches such as ensemble models to incorporate a larger set of attributes and enable more robust investigations of attribute changes on expected crash rates; utilizing the system’s scalability to incorporate yearly models; and incorporating real-time monitoring of crashes into the system to enable notification to practitioners when current crash rates exceed predicted values so potential problems can be addressed in a timely fashion.

Publication Date: 
Saturday, June 30, 2018
Publication Number: 
WA-RD 886.1
Last modified: 
02/27/2019 - 16:35
Jeremy Blum, Shuaiqi Huang, Puttipan Seraneeprakarn, Narayan Venkataraman, Venky Shankar.
Pennsylvania State University. Thomas D. Larson Pennsylvania Transportation Institute.
Number of Pages: 
Demonstration projects, Web applications, Mathematical models, Implementation, Highways, Crash risk forecasting, Types of roads by network, Highway factors in crashes, System safety, Strategic planning, Recommendations.