@Article{cmes.2025.063819,
AUTHOR = {Mohsen Ali Alawami, Dahyun Jung, Yewon Park, Yoonseo Ku, Gyeonghwan Choi, Ki-Woong Park},
TITLE = {DriveMe: Towards Lightweight and Practical Driver Authentication System Using Single-Sensor Pressure Data},
JOURNAL = {Computer Modeling in Engineering \& Sciences},
VOLUME = {143},
YEAR = {2025},
NUMBER = {2},
PAGES = {2361--2389},
URL = {http://www.techscience.com/CMES/v143n2/61440},
ISSN = {1526-1506},
ABSTRACT = {To date, many previous studies have been proposed for driver authentication; however, these solutions have many shortcomings and are still far from practical for real-world applications. In this paper, we tackle the shortcomings of the existing solutions and reach toward proposing a lightweight and practical authentication system, dubbed DriveMe, for identifying drivers on cars. Our novelty aspects are ① <i>Lightweight scheme</i> that depends only on a single sensor data (i.e., pressure readings) attached to the driver’s seat and belt. ② <i>Practical evaluation</i> in which one-class authentication models are trained from only the owner users and tested using data collected from both owners and attackers. ③ <i>Rapid Authentication</i> to quickly identify drivers’ identities using a few pressure samples collected within short durations (1, 2, 3, 5, or 10 s). ④ <i>Realistic experiments</i> where the sensory data is collected from real experiments rather than computer simulation tools. We conducted real experiments and collected about 13,200 samples and 22,800 samples of belt-only and seat-only datasets from all 12 users under different settings. To evaluate system effectiveness, we implemented extensive evaluation scenarios using four one-class detectors One-Class Support Vector Machine (OCSVM), Local Outlier Factor (LOF), Isolation Forest (IF), and Elliptic Envelope (EE), three dataset types (belt-only, seat-only, and fusion), and four different dataset sizes. Our average experimental results show that the system can authenticate the driver with an F1 score of 93.1% for seat-based data using OCSVM classifier, an F1 score of 98.53% for fusion-based data using LOF classifier, an F1 score of 91.65% for fusion-based data using IF classifier, and an F1 score of 95.79% for fusion-based data using EE classifier.},
DOI = {10.32604/cmes.2025.063819}
}