Classification of Driver Distraction: A Comprehensive Analysis of Feature Generation, Machine Learning, and Input...

Area of Science:

• Machine learning applications in transportation safety.
• Driver behavior analysis and prediction.
• Human-computer interaction in vehicles.

Background:

• Distracted driving significantly contributes to vehicle accidents, injuries, and fatalities.
• Increasing use of mobile devices and in-vehicle systems heightens the need for distraction detection.
• Developing effective driver distraction detection is crucial for road safety.

Purpose of the Study:

• To identify optimal machine learning algorithms for detecting driver distraction.
• To predict the sources of driver distraction using performance and physiological data.
• To evaluate advanced feature generation techniques for distraction detection.

Main Methods:

• Trained 21 machine learning algorithms on driver distraction data.
• Utilized physiological and driving behavioral inputs.
• Employed Time Series Feature Extraction based on Scalable Hypothesis tests (TSFX) for feature generation.

Main Results:

• A Random Forest algorithm, using only driving behavior data, achieved the highest performance in classifying driver distraction.
• Key input measures included lane offset, speed, and steering.
• Most effective feature types were standard deviation, quantiles, and nonlinear transforms.

Conclusions:

• Ensemble machine learning algorithms trained on driving behavior and nonstandard features can improve distraction detection.
• New indicators of distraction were identified from speed and steering measures.
• Future distraction mitigation systems should prioritize driver behavior-based algorithms with advanced feature generation.