Predicting microscopic vehicle collision risks at toll plaza diverging area using bayesian dynamic logistic...

Area of Science:

• Traffic Engineering
• Transportation Safety
• Data Science

Background:

• Toll plaza diverging areas without lane markings are high-risk traffic bottlenecks due to vehicle weaving.
• Existing conflict prediction models fail to adapt to dynamic traffic conditions.

Purpose of the Study:

• To develop a self-adaptive Bayesian dynamic logistic regression approach for predicting vehicle collision risks at toll plaza diverging areas.
• To improve the adaptability and efficiency of traffic conflict prediction models.

Main Methods:

• Extracted aggregated traffic characteristics from high-precision vehicle trajectory data.
• Utilized Extended Time-to-Collision (ETTC) to measure multi-directional collision risks.
• Developed Bayesian dynamic logistic regression models with varying data sampling strategies.

Main Results:

• Bayesian models demonstrated strong self-adaptive correction capabilities, with Area Under the Curve (AUC) values exceeding 0.9.
• Identified more influencing factors than standard logistic regression and required only 20% of the data for initialization.
• Continuous updating with incoming data significantly reduced computational demands.

Conclusions:

• The proposed Bayesian dynamic logistic regression approach offers enhanced predictive accuracy and efficiency for traffic collision risks.
• Incorporating richer prior information and continuous data updates improves model performance.
• Findings support the development of targeted management strategies for toll plaza safety.