Advanced

Predictive models for accidents on urban links - A focus on vulnerable road users

Jonsson, Thomas LU (2005) In Bulletin / Lund Institute of Technology, Department of Technology and Society 226.
Abstract
Much of earlier work on predictive models for accidents has been focused on rural traffic or urban intersections. This work has aimed at identifying and investigating possible improvements to predictive models for accidents on urban links. A special focus has been on the accidents of vulnerable road users.



The possible improvements investigated have been: a) the use of exposure data for vulnerable road users, b) the use of actual vehicle speeds, c) to separate vehicle accidents into single vehicle and multiple vehicle accidents and model each type separately.



The study involved eight Swedish cities, of which six were included in the development of accident models. In addition to police-reported traffic... (More)
Much of earlier work on predictive models for accidents has been focused on rural traffic or urban intersections. This work has aimed at identifying and investigating possible improvements to predictive models for accidents on urban links. A special focus has been on the accidents of vulnerable road users.



The possible improvements investigated have been: a) the use of exposure data for vulnerable road users, b) the use of actual vehicle speeds, c) to separate vehicle accidents into single vehicle and multiple vehicle accidents and model each type separately.



The study involved eight Swedish cities, of which six were included in the development of accident models. In addition to police-reported traffic accidents, data for the models were collected in specific field studies and partly from the cities.



The study shows that the inclusion of exposure data for vulnerable road users in the models for vulnerable road users? accidents greatly improves the predictive ability of the models. Vehicle speeds were found to be very difficult to use in the models as vehicle speeds are highly correlated with most of the other variables and make the model coefficients unstable. The separate modelling of single and multiple vehicle accidents failed partly as the single vehicle accidents were too few for constructing sound models. The models for multiple vehicle accidents developed, however, had a better predictive ability than models for all vehicle accidents. Land use was the next important explanatory variable for most accident types after the exposure variables.



Accident models were developed with one data set and validated with another data set. The models performed very well in prediction by explaining between 71% and 81% of the systematic variation in the validation data. The validation indicated that exponents were 0.5 for both the flows of pedestrians and motor vehicles in models for accidents involving vulnerable road users, and 1.0 for the motor vehicle flow exponent in the models for motor vehicle accidents. For bicyclist accidents the correct exponent for bicyclist flows is likely to be somewhat lower than 0.5, close to 0.35.



The study also recommended practical uses for the models, listed the lessons learned from the modelling as well as proposed a number of topics for further research. (Less)
Please use this url to cite or link to this publication:
author
supervisor
opponent
  • Professor Maher, Mike, Napier University, Edinburgh, UK
organization
publishing date
type
Thesis
publication status
published
subject
keywords
Road transport technology, Transportteknik, Transport technology, Teknik, Technological sciences, Stads- och glesbygdsplanering, Town and country planning, speed, links, urban, safety, vulnerable road users, modelling, prediction, crash, traffic safety, accident, Vägtransportteknik
in
Bulletin / Lund Institute of Technology, Department of Technology and Society
volume
226
pages
142 pages
publisher
Department of Technology and Society, Lund University
defense location
Room V:B, V-Building John Ericssons väg 1, Lund Institute of Technology
defense date
2005-03-04 13:15
ISSN
1404-272X
language
English
LU publication?
yes
id
df478570-7b0f-44ed-a5a5-19dbd8e9df4f (old id 24269)
date added to LUP
2007-05-31 15:38:16
date last changed
2016-09-19 08:44:53
@phdthesis{df478570-7b0f-44ed-a5a5-19dbd8e9df4f,
  abstract     = {Much of earlier work on predictive models for accidents has been focused on rural traffic or urban intersections. This work has aimed at identifying and investigating possible improvements to predictive models for accidents on urban links. A special focus has been on the accidents of vulnerable road users.<br/><br>
<br/><br>
The possible improvements investigated have been: a) the use of exposure data for vulnerable road users, b) the use of actual vehicle speeds, c) to separate vehicle accidents into single vehicle and multiple vehicle accidents and model each type separately.<br/><br>
<br/><br>
The study involved eight Swedish cities, of which six were included in the development of accident models. In addition to police-reported traffic accidents, data for the models were collected in specific field studies and partly from the cities.<br/><br>
<br/><br>
The study shows that the inclusion of exposure data for vulnerable road users in the models for vulnerable road users? accidents greatly improves the predictive ability of the models. Vehicle speeds were found to be very difficult to use in the models as vehicle speeds are highly correlated with most of the other variables and make the model coefficients unstable. The separate modelling of single and multiple vehicle accidents failed partly as the single vehicle accidents were too few for constructing sound models. The models for multiple vehicle accidents developed, however, had a better predictive ability than models for all vehicle accidents. Land use was the next important explanatory variable for most accident types after the exposure variables.<br/><br>
<br/><br>
Accident models were developed with one data set and validated with another data set. The models performed very well in prediction by explaining between 71% and 81% of the systematic variation in the validation data. The validation indicated that exponents were 0.5 for both the flows of pedestrians and motor vehicles in models for accidents involving vulnerable road users, and 1.0 for the motor vehicle flow exponent in the models for motor vehicle accidents. For bicyclist accidents the correct exponent for bicyclist flows is likely to be somewhat lower than 0.5, close to 0.35.<br/><br>
<br/><br>
The study also recommended practical uses for the models, listed the lessons learned from the modelling as well as proposed a number of topics for further research.},
  author       = {Jonsson, Thomas},
  issn         = {1404-272X},
  keyword      = {Road transport technology,Transportteknik,Transport technology,Teknik,Technological sciences,Stads- och glesbygdsplanering,Town and country planning,speed,links,urban,safety,vulnerable road users,modelling,prediction,crash,traffic safety,accident,Vägtransportteknik},
  language     = {eng},
  pages        = {142},
  publisher    = {Department of Technology and Society, Lund University},
  school       = {Lund University},
  series       = {Bulletin / Lund Institute of Technology, Department of Technology and Society},
  title        = {Predictive models for accidents on urban links - A focus on vulnerable road users},
  volume       = {226},
  year         = {2005},
}