Risky lane-changing behavior recognition based on stacking ensemble learning on snowy and icy surfaces

Area of Science:

• Traffic Safety Engineering
• Artificial Intelligence in Transportation
• Environmental Psychology in Driving

Background:

• Risky lane-changing (LC) behavior significantly compromises traffic safety, particularly on hazardous snowy and icy road surfaces.
• Existing research on risky lane-changing behavior (RLCB) in extreme weather conditions is limited due to data scarcity and surface complexities.
• Developing effective RLCB identification methods is crucial for enhancing vehicle safety in adverse conditions.

Purpose of the Study:

• To establish a novel research framework for identifying RLCB on highways under snowy and icy conditions.
• To select key risk characterization indicators (RCIs) for RLCB.
• To develop and evaluate a high-accuracy RLCB recognition model using driving simulation data.

Main Methods:

• A highway lane-changing scenario was simulated on snowy and icy surfaces, generating 1200 data samples.
• The C4.5 decision tree algorithm and Pearson correlation analysis were employed to select 12 key RCIs based on importance and low inter-correlation.
• A Stacking ensemble learning method was utilized to develop the RLCB recognition model, which was then compared against traditional algorithms.

Main Results:

• Twelve key RCIs were identified, balancing parameter importance and minimizing inter-correlation.
• The Stacking ensemble learning model demonstrated superior performance compared to traditional recognition algorithms.
• The developed RLCB recognition model achieved a high accuracy rate of 98.33%.

Conclusions:

• The study successfully developed a robust framework for identifying RLCB on snowy and icy surfaces.
• The Stacking ensemble learning model offers a significant advancement in RLCB detection accuracy.
• Findings provide a foundation for developing advanced lane-changing warning systems for intelligent connected vehicles operating in extreme weather conditions.