Using multivariate adaptive regression splines (MARS) to develop crash modification factors for urban freeway...

Area of Science:

• Transportation Engineering
• Traffic Safety
• Data Mining

Background:

• Crash modification factors (CMFs) are crucial for evaluating geometric design changes on roadways.
• Previous studies often overlooked the unique conditions of interchange influence areas on urban freeways.
• The Highway Safety Manual (HSM) and SafetyAnalyst tool have increased the demand for CMFs in the US.

Purpose of the Study:

• To develop CMFs for interchange influence areas on urban freeways in Florida.
• To analyze the impact of median width, inside shoulder width, and outside shoulder width on crash frequency.
• To investigate CMFs for total, fatal and injury (FI), rear-end, and sideswipe crashes.

Main Methods:

• Application of multivariate adaptive regression splines (MARS), a data mining technique.
• MARS was used to model nonlinear relationships and simultaneous effects of geometric variables.
• Comparison of MARS model performance against the negative binomial model using four years of crash data (2007-2010).

Main Results:

• MARS demonstrated superior prediction capability and goodness-of-fit compared to the negative binomial model.
• Key influential variables identified: outside shoulder width, inside shoulder width, median width, lane width, traffic volume, and shoulder type.
• A 2-ft increase in outside shoulders (10ft to 12ft) reduced FI crashes by 10%; a similar increase in inside shoulders reduced FI crashes by 33%.

Conclusions:

• Wider shoulders are associated with a reduction in fatal and injury crashes.
• Narrower medians significantly increase the frequency of rear-end, total, and FI crashes.
• MARS is an effective method for developing CMFs in complex urban freeway environments.