Collaborative driving style classification method enabled by majority voting ensemble learning for enhancing...

Area of Science:

• Transportation Science
• Machine Learning
• Traffic Safety Engineering

Background:

• Driver behavior analysis is crucial for traffic safety, but current methods face challenges in accuracy and stability.
• Existing driving style classification models often struggle with large parameter sets and inconsistent results.

Purpose of the Study:

• To develop an accurate and efficient collaborative driving style classification method.
• To investigate differences in driving behaviors among various vehicle drivers.
• To improve upon existing methods for autonomous driving and usage-based insurance.

Main Methods:

• A two-stage approach involving pre-classification and classification using ensemble learning.
• Utilizing composite clustering algorithms for initial data labeling (aggressive, stable, conservative).
• Employing a majority voting ensemble with three classifiers for classifying unlabeled data.

Main Results:

• The proposed collaborative method demonstrated good and stable performance in driving style classification.
• Achieved significant improvements in evaluation metrics: accuracy (+1.49%), precision (+2.90%), recall (+5.32%), and F-measure (+4.49%) compared to other methods.
• Outperformed similar classification techniques in overall performance.

Conclusions:

• The collaborative driving style classification method offers superior accuracy and efficiency.
• This approach is highly suitable for applications in autonomous driving and usage-based insurance.
• The method provides a robust solution for analyzing and differentiating driver behaviors.