A Recognition Method of Aggressive Driving Behavior Based on Ensemble Learning.

Area of Science:

• Traffic Safety
• Machine Learning
• Behavioral Analysis

Background:

• Aggressive driving behavior (ADB) is a significant factor in traffic accidents.
• Existing data-driven ADB recognition methods suffer from high miss rates and low accuracy due to imbalanced datasets and single classifiers.
• Accurate ADB recognition is crucial for timely driver warnings and interventions.

Purpose of the Study:

• To propose an ensemble learning-based method for accurate recognition of aggressive driving behavior (ADB).
• To address limitations of previous ADB recognition methods, including imbalanced class distribution and single classifier usage.
• To enhance the performance of ADB detection systems for improved road safety.

Main Methods:

• Employed self-organizing map (SOM) for grouping majority classes to create balanced datasets.
• Utilized deep learning models—Convolutional Neural Networks (CNN), Long Short-Term Memory (LSTM), and Gated Recurrent Unit (GRU)—as base classifiers.
• Combined base classifiers using 10 different ensemble rules and validated on a multi-source naturalistic driving dataset.

Main Results:

• The proposed ensemble learning method significantly outperformed typical deep learning methods in ADB recognition accuracy, recall, and F1-score.
• The optimal ensemble classifier combined LSTM with the Product Rule.
• An LSTM-based ensemble using the Sum Rule demonstrated suboptimal but still improved performance.

Conclusions:

• Ensemble learning, particularly with LSTM and the Product Rule, offers a superior approach for recognizing aggressive driving behavior.
• The developed method effectively handles imbalanced datasets and improves the reliability of ADB detection systems.
• This research contributes to developing more effective driver assistance systems for enhancing traffic safety.