Development of Crash Prediction Models for Short-Term Durations (2023)

Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

175   Web-Based Tool, User-Friendly Practitioners’ Guide, and Training Materials The research team prepared a user-friendly guide that aims to help safety analysts use the developed short-term crash prediction models in practice. It discusses when and how the alternative methods should be selected and implemented based on local data availability, data quality, and expertise in implementation. Further, it provides the step-by-step process to prepare the dataset and utilizes the short-term SPFs for the different crash severity levels and time periods. This user guide was submitted as a separate document. In addition to this user guide, the team developed a web tool for the short-term crash predic- tion of different use case scenarios. However, the user should prepare the datasets by following the provided steps in this user guide before utilizing the web tool. The user guide also includes snapshots, descriptions, and guidance on how to use the web tool. The team conducted multiple tests on the web-based tool using samples from the freeway segment types and use case scenarios, including urban freeway segment type, VSL/VAS, HOV lanes, and weaving type. Figure 52 illustrates an example of the test results. It illustrates the observed and predicted crashes using the tool for the KABCO AM-peak weaving type A segments. The URL for the web-based tool is https://www.highwaysafetymanual.org/Pages/Tools.aspx. The proposed approach and developed short-term crash prediction models were presented to an expert group to obtain their feedback from the practitioners’ point of view. The team invited experts from state DOTs, transportation industry, and academia. The expert group included: 1. James Bonneson, PhD Senior Principal at Kittelson & Associates, Inc. 2. Alan El-Urfali, PE State Traffic Services Program Engineer at Florida Department of Transportation Substituted by FDOT Safety Engineer Dr. Dibakar Saha 3. Pete Jenior, PE, PTOE Associate Engineer at Kittelson & Associates, Inc. 4. Chris Lee, PhD, PE Associate Professor in Civil and Environmental Engineering at University of Windsor; Associate Editor for Accident Analysis and Prevention; Member of Transportation Division of Canadian Society for Civil Engineering (CSCE) Technical Committee 5. Xiao Qin, PhD Professor in Civil and Environmental Engineering Department at University of Wisconsin Milwaukee; Director of the Institute for Physical Infrastructure and Transportation (IPIT); Founder and Director of Safe and Smart Traffic Lab 6. Stephen Read, PE Highway Safety Planning Manager at Virginia Department of Transportation C H A P T E R 8

176 Development of Crash Prediction Models for Short-Term Durations 7. Raghavan Sirinivasan, PhD Senior Transportation Research Engineer at the University of North Carolina Highway Safety Research Center The team presented and discussed the models’ development, calibration, and evaluation with the expert group. The team considered all the comments and recommendations from the expert group while working on the project. The discussion and recommendations were as follows: • The expert group supported and complimented the extensive work and the research idea of developing short-term crash prediction models for different time periods and use case scenarios. • The expert group suggested considering the natural logarithm of different variables other than the volume while developing the models. The team already utilized the natural loga- rithm of average speed in the developed models. Further, the team considered including the natural logarithm of the standard deviation of speed; however, they found it to be insignificant in the developed models. • The expert group discussed adding geometric variables and the comparison between the basic models with and without geometric variables. The expert group supported the team’s conclusion that the short-term traffic parameters, especially the speed and uniformity of speed, could capture much of the driving behavior in response to the geometry. • The expert group suggested adding geometric variables for the left side of the roadway seg- ments (i.e., left shoulder width, median type), the grades, and the curvature of the roadway segment. However, due to the availability of the data and the sample size, the team could not include these variables in the developed models. • The expert group suggested checking whether the geometry could play a role in specific use case scenarios. The team already considered including the geometric variables in some cases. However, the performance of the models did not improve much when considering them. Hence, for simplicity and availability of the data, the team suggested utilizing the models without geometric variables. • The expert group suggested including the truck percentages in the models. However, the high-resolution data of the truck percentages were mostly unavailable for the utilized states. 0 5 10 15 20 25 30 35 1 6 11 16 21 26 31 36 41 46 51 56 61 66 71 76 81 86 91 96 10 1 10 6 11 1 11 6 12 1 12 6 13 1 13 6 14 1 14 6 15 1 N um be r o f C ra sh es Segments Testing the Web-Based Tool - Weaving Segments Type A (AM Peak) Total Crashes Predicted Crashes Figure 52. Observed and predicted crashes web-based tool testing (Weaving Type A).

Web-Based Tool, User-Friendly Practitioners’ Guide, and Training Materials 177   • The expert group supported utilizing states’ dummy variables to tune the coefficients of the traffic variables in the model. Further, for implementation purposes, the group supported adding the calibration factor for the generic models so that it could be utilized by other states. • The expert group supported the evaluation approach that the research team utilized using the absolute error for different accuracy levels. The group also mentioned that the accuracy was acceptable. • The correlation between different traffic parameters should be checked while developing the short-term models. The team already checked the correlation before including the variables in the model. • The expert group recognized that the developed short-term models could be useful in the safety evaluation of future conditions by including the micro-simulation output variables in the developed models. • Separate models that include segments with multiple ATM strategies should be considered. The team will consider looking into the combination of the ATM strategies after finalizing the different models for the use case scenarios. • The team should consider developing models for KAB because they are more practical to use from the practitioners’ point of view. Hence, the team started developing the KAB models for all the use case scenarios.