Summer Program on Transportation Statistics, Statistical Challenges for Advances in Estimation of Traffic Exposure and Crash Prediction - John N. Ivan, Aug 14, 2017
Crashes on limited access roadways typically occur due to drivers being unable to react in time to avoid collisions with vehicles ahead of them either moving slower or merging
unexpectedly. Prevailing traffic stream conditions with high volume and low or variable speed downstream of low volume and high speed conditions can increase the possibilities for such collisions to occur. Real time trajectories of vehicles collected through crowd sourcing methods can give information about the distribution of speeds in the traffic stream by space
and time. Spatio-temporal models relating these observed speed distributions to the occurrence of crashes or near crashes can help to identify crash prone traffic conditions as
they arise, offering the opportunity to warn drivers before crashes occur.
CLIM Undergraduate Workshop: Statistical Development and challenges for Paleo...
Similar to Summer Program on Transportation Statistics, Statistical Challenges for Advances in Estimation of Traffic Exposure and Crash Prediction - John N. Ivan, Aug 14, 2017
Similar to Summer Program on Transportation Statistics, Statistical Challenges for Advances in Estimation of Traffic Exposure and Crash Prediction - John N. Ivan, Aug 14, 2017 (20)
Summer Program on Transportation Statistics, Statistical Challenges for Advances in Estimation of Traffic Exposure and Crash Prediction - John N. Ivan, Aug 14, 2017
1. STATISTICAL CHALLENGES FOR
ADVANCES IN ESTIMATIONOF
TRAFFIC EXPOSUREAND CRASH
PREDICTION
Summer Program on
Transportation Statistics: August 14, 2017
John N. Ivan, University of Connecticut
2. Three Challenges inTransportation
Engineering
Topic 1
• MissingValues
inTraffic
Counts
Topic 2
• Driver, vehicle
and
environmental
data for
roadway
crash
prediction
Topic 3
• Microscopic or
individual
safety analysis
4. Traffic engineering applications require vehicular
volumes at various aggregation levels
Annual Average Daily
Traffic (veh/day)
• Safety evaluations
• Planning
Hourly turning
volume or link counts
(veh/h)
• Traffic signal timing
• Traffic level of
service computation
Classification by
vehicle type (heavy
vehicle percentage)
• Pavement design
• Level of service
computation
5. Traffic counts are taken at limited scope
in space and time: AADT
Source: www.JamarTech.com
■ Observed automatically
– Tubes or Inductive Loop Detectors
– 24 hour count once every 3 years
■ Extrapolated to annual average
– Expansion factors categorized by functional
classification and rural/urban
– Ignores temporal variation within each
classification and rural/urban category
■ Hourly distribution is ignored
– observed but not considered in AADT
6. Hourly turning volume or link counts
Source: www.JamarTech.com
■ Observed manually
– human observers
■ Irregular intervals
– Usually for a specific project
■ Peak periods
– morning and afternoon on
one day
7. Vehicle classification (heavy vehicle)
counts
■ Vehicle classifications are
observed either manually or
automatically and checked
manually
■ Taken at permanent count
stations or sampled locations
across the state
■ Not always taken at all
locations Source: pixabay.com
8. Continuous count stations
■ Hourly counts are taken
continuously using
inductive loop detectors
■ Used to derive expansion
factors for coverage
counts
■ Can be missing values
when the detector
malfunctions
Source: www.Flickr.com
9. Statistical innovation opportunities to
improve accuracy of traffic counts (1)
■ AADT
– Update extrapolation of AADT using coverage counts as
well as continuous counts
– Optimize scheduling of coverage counts
■ Hourly turning counts
– Augment “found” turning counts with additional
observations to adjust to an annual average or to predict
for any time of the year
10. Statistical innovation opportunities to
improve accuracy of traffic counts (2)
■ Classification counts
– Analyze existing classification counts and identify
spatial and temporal patterns that could help with
extrapolating one day classification counts to the full
year or another time or location
■ Continuous counts
– Impute missing hourly counts from the full time series of
counts; consider other counts in the network as well
12. Categories of
factors
contributing to
or associated
with crashes
Roadway also includes
environment
Venn diagram source:
https://www.fhwa.dot.gov/publications/pub
licroads/95winter/p95wi14.cfm
13. Only roadway
characteristics are
usually considered in
crash prediction models
for segments and
intersections
Environment, driver and vehicle
characteristics vary by time at
each analysis location
We are reaching the limits of
crash prediction model accuracy
improvement without
considering these other factors
14. Maturing and
emerging
technologies offer
new opportunities
for providing data
to fill this gap
Crowdsourced
data
Driver and vehicle
demographics
Individual travel patterns
purchased from cellular
carriers or collected
voluntarily
Proper statistical sampling
adjustments will need to be
derived and applied
Weather
sensors
Rainfall and temperature
data
Distributed geographically
and not located at every
road location
Spatial and temporal
modeling would be
required to translate from
point to roadway locations
16. Disaggregated crash prediction approach
■ Cases are individual vehicles (and occupants) traversing each
segment or intersection, rather than road segments or
intersections
■ Combines the crash count prediction problem with the crash
severity prediction problem – possible by doing each at the
same level of aggregation
■ Input variables could include environmental, driver and
person information as well as roadway characteristics
■ Linking EMS, hospital and DMV data would be helpful for
including additional variables that could be associated with
the crash outcome.
17. Three stage modeling process requiring
hierarchical statistical methodologies
Probability of an individual
driver or drivers resulting in
a type of crash
• Roadway characteristics
• Driver experience,
impairment
• Vehicle condition, crash
avoidance features
• Road surface condition,
visibility
Probability of the extent of
damage to each vehicle
involved
• Type of crash
• Vehicle speeds before
and at time of crash
• Mass and configuration
of each vehicle
Probability of the level of
severity for each individual
involved
• Extent of damage to
vehicle
• Seating position
• Age, sex, physical
condition