Developing crash prediction models using parametric and nonparametric approaches for rural mountainous freeways: A...

Area of Science:

• Transportation Engineering
• Traffic Safety Analysis
• Statistical Modeling

Background:

• Wyoming's I-80 is a critical freight route facing severe winter conditions and high fatality rates.
• The U.S. Department of Transportation (USDOT) is piloting connected vehicle technology on this corridor to enhance safety and mobility for heavy trucks.
• Pre-deployment safety performance evaluation is crucial for quantifying the effectiveness of this pilot program.

Purpose of the Study:

• To conduct a comprehensive safety performance evaluation of the entire 402-mile I-80 corridor in Wyoming.
• To develop and compare Safety Performance Functions (SPFs) using various statistical models.
• To inform the effectiveness assessment of connected vehicle technology deployment.

Main Methods:

• Homogeneous roadway segmentation based on geometric characteristics.
• Transferability analysis to assess representativeness of shorter corridor segments.
• Development and comparison of crash prediction models: Negative Binomial (NB), Spatial Autoregressive (SAR), and Multivariate Adaptive Regression Splines (MARS).

Main Results:

• The entire 402-mile I-80 corridor requires analysis due to significant variations.
• The MARS model demonstrated a superior fit for crash prediction compared to NB and SAR models (indicated by lower AIC values).
• SAR models identified significant spatial dependency, while NB models were superior when spatial correlation was not significant.

Conclusions:

• The MARS model is effective for developing SPFs, offering a better fit than NB and SAR models in this context.
• Spatial dependency should be considered in crash modeling, with SAR models being appropriate when significant.
• A combination of parametric and non-parametric modeling techniques is recommended for developing robust SPFs based on specific analytical needs.