A Machine-Learning Approach to Distinguish Passengers and Drivers Reading While Driving.

Area of Science:

• Human-computer interaction
• Machine learning applications in transportation safety
• Mobile sensing technologies

Background:

• Driver distraction from smartphone use is a significant cause of traffic accidents.
• Increased connectivity and social media use exacerbate the problem of in-vehicle smartphone distraction.
• Current methods for detecting distracted driving are often intrusive or limited.

Purpose of the Study:

• To propose a non-intrusive technique for distinguishing drivers from passengers using only smartphone sensor data.
• To apply machine learning to automatically detect distracted driving behavior related to smartphone use.
• To evaluate the effectiveness of various machine learning models in this detection task.

Main Methods:

• Utilized smartphone sensors (e.g., accelerometer, gyroscope) to collect data during driving scenarios.
• Modeled and evaluated seven advanced machine learning techniques, including Convolutional Neural Networks (CNN) and Gradient Boosting.
• Tested the models across diverse driving scenarios to assess their robustness.

Main Results:

• Convolutional Neural Networks (CNN) and Gradient Boosting demonstrated superior performance in distinguishing drivers from passengers.
• All evaluated metrics, including accuracy, precision, recall, F1-score, and kappa, exceeded 0.95.
• The proposed non-intrusive technique achieved high effectiveness in real-world scenarios.

Conclusions:

• Smartphone sensor data combined with machine learning offers a promising non-intrusive solution for detecting distracted driving.
• The developed method can accurately differentiate between drivers and passengers, aiding in safety interventions.
• This approach has the potential to significantly improve road safety by identifying and mitigating smartphone-induced driver distraction.