Rural Speed Safety Project for USDOT SDI
Subasish Das, Ph.D.
December 7, 2018
Defining the Research Space
Research questions:
- How do human factors become dominant in crash severities?
- How does a pedestrian’s body dynamics effect the driver?
- Do the association rules identify any domain specific relationship?
Challenges
- Speed data- quantity, usability
- Two linear systems- crash/roadway geometry/traffic volume data and speed data
- Data structure for the evaluation
- Excess number of rural sites with zero crashes
Challenge= Speed data- quantity, usability
- Solution= NPMRDS data are available to DOTs / able to explore greater number of speed measures
Challenge= Two linear systems- crash/roadway geometry/traffic volume data and speed data
- Conflation of two linear systems
Challenge= Data structure for the evaluation
- Create 3 data structures where Data Structure 1 and 2 will address Q1 and Data Structure 3 will addrsess Q2
Challenge= Excess number of rural sites with zero crashes
- Develop best-fit statistical models to address zero crashes
- Currently used models are : negative binomial (NB), NB Lindley (NBL), and random parameter NBL
Evaluation
- Exploratory data analysis
- Deveoping models by facility type
- Predicted crashes= function (segment length, traffic volume, other imporatant geometry, and speed measures)
- This project can inform Highway Safety Manual (HSM) models by incorporating speed measures
- Develop expected crashes
- Expected crashes= Incorporate weighting factor as described in the HSM
Statistical Analysis Findings
- Q1: Do different speed measures contribute to crash outcomes?
- Yes!
- Developed decision support tool to illustrate number of expected crashes on segments
- Q2: Is there more variability in speeds just prior to a crash?
- Yes!
- Developed models that use travel speed to classify crash and non-crash event
Prototype of Decision Support Tool
Live app link: https://ldwhite.shinyapps.io/RuralSpdSafety_Demo/
Next Steps by Questions
- Q1: Do different speed measures contribute to crash outcomes?
- Q2: Is there more variability in speeds just prior to a crash?
Data Preparation
- Three States: North Carolina, Washington, and Ohio
- Three Databases
- Highway Safety Information System (HSIS)
- Highway Performance Monitoring System (HPMS)
- National Performance Management Research Data Set (NPMRDS)
- Year: 2015
- Task
- Conflate 3 databases for 3 states
SPFs at Segment level (Washington)
SPFs at Segment-temporal level (Washington)
Proximity of Crashes at Segment-temporal Level