Highway-rail grade crossings accident prediction using Zero Inflated Negative Binomial and Empirical Bayes method

Area of Science:

• Transportation Safety
• Statistical Modeling
• Road Safety Engineering

Background:

• The Federal Railroad Administration (FRA) developed a new accident prediction model for railroad grade crossings.
• This model utilizes a Zero Inflated Negative Binomial (ZINB) with Empirical Bayes (EB) adjustments.
• A limitation of the FRA model is its single equation for all warning devices, potentially misrepresenting field conditions.

Purpose of the Study:

• To introduce and evaluate the ZINEBS model for railroad grade crossing accident prediction.
• To address the limitations of the current FRA model by incorporating device-specific variables.
• To improve the accuracy of accident frequency predictions based on warning device types.

Main Methods:

• Development of the Zero Inflated Negative Binomial with Empirical Bayes adjustment System (ZINEBS) model.
• Implementation of three distinct equations within ZINEBS, tailored to specific warning devices (gates, flashing lights, crossbucks).
• Comparison of ZINEBS model predictions against field data and the existing FRA model.

Main Results:

• The ZINEBS model provides predictions that more closely align with field data compared to the FRA model.
• This improved accuracy was consistent across all three analyzed warning device types.
• The study highlights that device-specific variables enhance prediction accuracy.

Conclusions:

• The ZINEBS model offers a more accurate approach to predicting railroad crossing accidents than the current FRA model.
• ZINEBS complements the FRA model, particularly when specific traffic control device characteristics are crucial.
• The findings suggest that accounting for device-specific variables leads to better safety predictions.