A Machine Learning-Based Framework for Risk Recognition and Reliability Evaluation in City Expressway Ramp Merging.

Area of Science:

• Traffic Engineering
• Machine Learning
• Road Safety

Background:

• Ramp merging poses significant safety risks due to complex vehicle interactions.
• Existing methods often suffer from 'future information' leakage, compromising accurate risk assessment.

Purpose of the Study:

• To develop a novel framework for accurate and temporally unbiased risk identification in ramp merging scenarios.
• To improve the reliability and stability of dynamic traffic risk assessment.

Main Methods:

• An 'observation-conflict' mechanism was employed to decouple risk identification processes.
• Kinematic features were extracted from the merging preparation phase for risk prediction.
• A three-level risk labeling scheme integrated critical time-to-collision (cpTTC) thresholds and duration constraints.

Main Results:

• The XGBoost model demonstrated superior performance with 0.8182 overall accuracy and 0.8898 multiclass AUC (OvR).
• The framework showed reliable statistical stability across varying traffic flow states.
• A risk recall of 0.9978 was achieved in a binary classification task.

Conclusions:

• The proposed framework offers a robust methodological basis for dynamic traffic risk assessment in ramp merging.
• The findings provide valuable early-warning features for enhancing road safety.
• The approach mitigates temporal bias and aligns risk detection with physical dynamics.