Trajectory data based freeway high-risk events prediction and its influencing factors analyses

Area of Science:

• Traffic Safety Engineering
• Transportation Science
• Data Science in Transportation

Background:

• Frequent traffic crashes cause significant global losses.
• Existing research often uses aggregated data, lacking microscopic insights into crash precursors.
• High-resolution traffic data is needed to understand detailed vehicle dynamics near potential crash sites.

Purpose of the Study:

• To analyze the relationship between microscopic traffic operation conditions and high-risk events.
• To develop and compare logistic regression models for predicting high-risk events.
• To identify key traffic parameters influencing the occurrence of high-risk events.

Main Methods:

• Utilized the HighD Dataset containing vehicle trajectory data from German highways.
• Identified high-risk events using Modified Time to Collision (MTTC) < 2 seconds.
• Extracted traffic operation characteristics within 5 seconds prior to events.
• Developed and compared standard logistic regression, random-effects logistic regression (RELR), and random-parameter logistic regression (RPLR) models.

Main Results:

• The random-parameter logistic regression (RPLR) model demonstrated superior fitness and prediction accuracy.
• Disturbed traffic flow (longitudinal and lateral) and excessively close following distances positively impact high-risk event occurrence.
• The RPLR model achieved 97% prediction accuracy for high-risk events 2 seconds in advance.

Conclusions:

• Microscopic traffic dynamics, particularly flow disturbances and vehicle proximity, are critical predictors of high-risk events.
• The RPLR model offers a robust tool for proactive traffic safety analysis and prediction.
• Findings can inform targeted safety countermeasures to mitigate crash likelihood.