Safer Roads Ahead: Machine Learning Solutions Spotting Distractions from Texting to Voice Messaging

By Di Yang & Mansoureh Jeihani

Distracted driving, particularly due to cellphone usage, is a growing concern in road safety. While texting has been widely studied as a distraction, voice messaging, despite its frequent use, has received far less attention. This study aims to fill that gap by focusing on how both texting and voice messaging contribute to distracted driving and how these behaviors can be detected using machine learning (ML) methods. The ultimate goal is to develop technology that can help reduce distracted driving and improve overall traffic safety.

To explore this, researchers collected data from 92 participants who used a driving simulator to navigate a simulated version of Baltimore’s metropolitan area. The driving simulator recorded various key aspects of driving behavior, such as speed, how often drivers used the brakes, throttle, steering wheel movement, when brake lights were activated, and how far the vehicle veered from the center of the road. These variables were crucial in building ML models designed to detect whether a driver was distracted by texting or voice messaging.

Several popular machine learning algorithms were tested on the data, including support vector machines (SVM), k-nearest neighbors (KNN), decision trees (DT), neural networks (NN), and adaptive boosting (AdaBoost). The purpose was to find the best model for detecting distracted driving events based on the driving performance data. The models were evaluated using a variety of performance metrics, including accuracy (how often the model correctly identified distractions), precision, sensitivity (how well the model detected actual distractions), the Matthews correlation coefficient (MCC), and the area under the receiver operating characteristic curve (AUC-ROC). The AUC-ROC is a critical measure that assesses how well the model distinguishes between distracted and non-distracted driving events.

The study found that the AdaBoost algorithm outperformed the others, achieving an accuracy of 74.67% and an AUC-ROC score of 82.5%. This means that AdaBoost was able to identify distracted driving events with a high degree of reliability and offered a strong balance between sensitivity (identifying distractions) and specificity (avoiding false positives). Neural networks and decision trees also performed moderately well, with accuracy scores of 72.04% and 70.1%, respectively. On the other hand, SVM and KNN models had lower accuracy, suggesting they were less effective in this context.

One of the most significant outcomes of this study is the potential for these machine learning models to be incorporated into real-world systems, such as in-vehicle warning technologies. These systems could monitor a driver’s behavior in real-time, detect when they are distracted by texting or voice messaging, and provide immediate alerts to help them refocus on the road. Such systems could be integrated as aftermarket products that drivers can add to their cars or even be built directly into vehicles by automakers. By detecting distractions early, these systems have the potential to reduce accidents and make driving safer.

In summary, this study demonstrates the potential of machine learning to detect distracted driving caused by texting and voice messaging. By utilizing vehicle trajectory data, the research lays the groundwork for developing effective in-vehicle warning systems that could significantly improve road safety. Although there are limitations to the current research, the findings represent an important step toward understanding and mitigating the dangers of distracted driving.

Read the Article

Article details

Text and Voice Message Distraction Detection: A Machine Learning Approach Using Vehicle Trajectory Data
Abolfazl Taherpour, Parisa Masoumi, Alireza Ansariyar, Di Yang, Samira Ahangari, and Mansoureh Jeihani
DOI: 10.1177/0361198124125359
First Published: June 24, 2024
Transportation Research Record: Journal of the Transportation Research Board

About the Authors

Featured

Dr. Di Yang is an Assistant Professor in the Department of Transportation & Urban Infrastructure Studies. His research focuses on developing statistical and data-driven methods at the intersection of transportation safety, connected and automated vehicles, human factors, policy analysis, and shared mobility. Dr. Yang's research contributions have resulted in over 40 publications in peer-reviewed journals, conference proceedings, and technical reports. His research has received funding from the National Science Foundation (NSF), U.S. Department of Transportation, among others. Dr. Yang serves on the Editorial Advisory Board of the journal Accident Analysis and Prevention, and is a Handling Editor for the journal Transportation Research Record.

Dr. Mansoureh Jeihani is a professor and the director of the SMARTER Regional UTC. She has over 20 years of experience in applied research in transportation and has over 140 publications. She has also been the PI/Co-PI for over 60 research grants totaling over $30M. She is the chair of Maryland Attainment Report Advisory Committee and the Maryland CAV Technical Subgroup; an executive member of the CUTC; a member of the TRB’s Artificial Intelligence and Advanced Computing Applications committee, and MDQI and DCQI Innovations Subcommittee. Dr. Jeihani was on the Safety panel of the White House ARPA-I Summit in 2023.