Improving crash predictability of the Highway Safety Manual through optimizing local calibration process

Area of Science:

• Transportation Engineering
• Traffic Safety
• Statistical Modeling

Background:

• The Highway Safety Manual (HSM) uses uncalibrated safety performance functions (SPFs) and crash modification factors (CMFs), requiring aggregate local calibration factors (LCFs).
• Current HSM calibration methods ensure overall prediction accuracy but compromise individual site prediction quality.
• Aggregate calibration limits the effectiveness of predictive methods for specific locations.

Purpose of the Study:

• To propose and evaluate an alternative method for locally calibrating the HSM predictive method.
• To improve crash prediction accuracy at individual locations while maintaining aggregate accuracy.
• To enhance the identification of road safety hot-spots and inform site-specific safety strategies.

Main Methods:

• Developed a methodology incorporating multiple calibration factors for SPF parameters and CMFs, unlike the HSM's single aggregate factor.
• Applied calibration factors using weight and power functions to better reflect local conditions.
• Estimated parameters using an optimization process with five different methods on rural two-lane, two-way (R2U) road data from Maryland, Illinois, and Washington, utilizing a developed tool (RSDI).

Main Results:

• The proposed local calibration method significantly outperformed the standard HSM calibration, calibrated models, and other alternative methods based on Goodness-of-Fit measures and CURE plots.
• Incorporating additional parameters for CMFs demonstrably improved prediction accuracy, overcoming limitations of previous studies.
• The methodology ensures calibration at the aggregate level while enhancing individual site prediction.

Conclusions:

• The proposed local calibration approach offers superior accuracy for individual site crash prediction compared to existing HSM methods.
• This enhanced prediction accuracy can lead to more effective identification of problem locations and targeted safety interventions.
• Further research is recommended to explore additional avenues for improving roadway safety predictive modeling.