Modelling Vehicle Accidents and Efficiency of Highway Geometric Design

Abstract

This study investigates the relationship and effect of geometric variables
such as curve radius (Horizontal and vertical), road width, median
width, shoulder width and super-elevation on the frequency of accident
in highways. Accident predictions were done for the Benin-Ore road and
Benin-Agbor road using field data such as numbers of crashes, length of
the route and Average Annual Daily Traffic (AADT) values etc. Poisson
and Negative Binomial Regressions techniques were adopted for
accident prediction modelling and model validation was done using the
Statistical Package for Social Sciences (SPSS) to determine the
reliability of the parameters. Where the Poisson regression model failed
because of over-dispersion and under-dispersion of data, the Negative
Binomial (NB) proved very useful because of its ability to cater to such
shortfall. Results from the analysis show that of the designated geometric
(independent) variables selected, vertical curves and shoulder-width
were the most significant in influencing the probability of crash
occurrence in the designated highways of the study domain. For the
Benin Ore road, the shoulder width and the road width accounted for the
principal causes of accident with a probability of crash occurrence of
73.6% and 8.7% respectively. The vertical alignment with a probability
of crash of 19.60% accounted for the principal cause of accident along
the Benin Agbor expressway. Super elevation and the horizontal curve
accounted for just 1% each of the probability of crash occurrence along
the Benin Agbor road. It can, therefore, be concluded that some form of
checks be carried out on the shoulder width and the road width along
the Benin Ore road while similar checks must be focused on the vertical
alignment along the Benin Agbor road to mitigate the occurrence of
accidents arising from defective geometric configuration.