Feature selection for driving style and skill clustering using naturalistic driving data and driving behavior...

Area of Science:

• Traffic Safety and Human Factors
• Machine Learning in Transportation

Background:

• Driving style and skill significantly impact traffic safety, efficiency, and capacity.
• Existing research often uses clustering algorithms for driving behavior analysis, but feature selection remains a challenge.

Purpose of the Study:

• To propose and validate a novel feature selection method for clustering driving behavior data.
• To identify key features that characterize individual driving styles and skills without prior labels.
• To improve the accuracy of driving behavior modeling and driver classification.

Main Methods:

• Developed a supervised machine learning model for driver identification using permutation importance for feature selection.
• Extracted 72 feature candidates from naturalistic driving data using 18 different methods.
• Validated feature selection using the Driving Behavior Questionnaire (DBQ) and analyzed driver group characteristics.

Main Results:

• Identified five key features: longitudinal acceleration, frequency centroid of longitudinal acceleration, shape factor of lateral acceleration, root mean square of lateral acceleration, and standard deviation of speed.
• Successfully clustered drivers into three distinct groups: novice, experienced cautious, and experienced reckless.
• Demonstrated the validity of the selected features through analysis of DBQ responses and feature distributions within each group.

Conclusions:

• The proposed feature selection method effectively identifies critical driving behavior indicators.
• The identified key features enable accurate classification of drivers based on their style and skill.
• This approach offers potential for enhanced driver characterization and improved driving behavior modeling for traffic safety.