Improvements in estimating a fatal accidents model formed by an Artificial Neural Network

Abstract

The Smeed Equation (SE) is the first model being used to improve the estimation of the number of dead in accidents that consist of the independent variables of population and number of vehicles and the dependent variable of the number of dead. In this study, the population variable in the SE is replaced with the number of drivers. At first, the SE is made suitable for USA data and the Revised SE is obtained. Then the coefficients are calculated again by the replacement of the number of drivers with population and the Improved SE is obtained. Afterwards, Artificial Neural Network (ANN) models are formed in both variable groups of population and number of drivers. The best ANN model, whose inputs are the number of vehicles and drivers, has 19 neurons, a tan-sig transfer function and a Levenberg—Marquardt training algorithm. In the comparison of ANN models and SE models, the value of R² increases from 0.8906 to 0.9695 and the value of mean square errors (MSEs) decreases from 87,503 to 39,310. As a result the replacement of the number of drivers variable with population has a contribution in the estimation of the number of dead in vehicle accidents. This study showed that use of the number of drivers instead of the population in the number of dead prediction can be improved with the accuracy of the proposed models. Moreover, ANN models can be used to predict the number of dead in traffic accidents with a high correlation coefficient and a low MSE according to the SE and loglinear regression methods.

Keywords

Artificial Neural Network fatality in traffic accidents motor-vehicle accident number of drivers

Get full access to this article

View all access options for this article.

References

World Health Organisation (WHO). World Report on Road Traffic Injury Prevention, Chapter 3 Risk Factors , Geneva, 2004, http://whqlibdoc.who.int/publications/2004/9241562609.pdf [accessed 10 May 2006].

Smeed RJ Some statistical aspects of road safety research. J Royal Stat Soc 1949; 112: 1-34.

Fouracre PR , Jacobs GD Further research on road accident rates in developing countries. TRRL Supplementary Report 270. Crowthorne: Transport Research Laboratory, 1977.

Leeming JJ Comparing international road accident deaths. AGARD Conf Proc 1976; 41(1761): 15-17.

Adams J. 1985. Smeed’s law, seat belts, and the emperor’s new clothes. Human Behaviour and Traffic Safety Conference, Plenum Press (GM Symposium Series), 1985, p.193-257.

Silvak M. Society’s aggression level as a predictor of traffic fatality rate. J Saf Res 1983; 14: 93-99.

Adams JGU . Smeed’s law: Some further thoughts . Traffic Eng Contr 1987; 10(7): 70-73.

Fieldwick R. , Brown RJ The effect of speed limits on road casualties . Traffic Eng Contr 1987; 28: 635-640.

Oppe S. Macroscopic models for traffic and traffic safety. Accident Anal Prev 1989; 21: 225-232.

10.

Jacobs GD , Palmer CJ Road safety in the emerging nations. Intertraffic Middle East ’96 Safety Symposium, 1996.

11.

Jacobs G. , Aeron-Thomas A. , and Astrop A. Estimating global road fatalities, TRL Report 445, ISSN 0968-4107, Copyright Transport Research Laboratory, 2000.

12.

Federal Highway Administration (FHWA) , U.S., Department of Transportation, http://www.fhwa.dot.gov/policy/ohpi/hss/index.htm [accessed 07 July 2006].

13.

Cybenko G. Approximation by superpositions of sigmoidal function. Math Control Signals Syst 1989; 2: 303-314.

14.

Hornik K. , Stinchcombe M. , and White H. Multilayer feedforward networks are universal approximators. Neural Network 1989; 2: 359-366.

15.

Matlab. Neural network toolbox. Natick, MA: The MathWorks Inc.