Fingerprinting drivers by applying Machine Learning on Raw CAN Bus Telemetry Data

Sammanfattning

The Controller Area Network (CAN) is a central part of modern vehicles because it enables communication between the different Electronic Control Units (ECUs) of the vehicle. This communication can be collected using On-Board Diagnostics II (OBD-II) requests, or by capturing all the unfiltered and encoded communication on the bus, referred to as raw CAN data. Previous studies have shown that this data can be used to fingerprint drivers by applying machine learning. The majority of work in this area has used datasets collected using the OBD-II approach, providing interpretable values such as engine speed, with the downside of being limited to the information provided by this protocol. To the best of our knowledge, no previous work has conducted fingerprinting using a publicly available dataset containing raw CAN data. A recent dataset that was published in November 2025 addresses the limitations in previous datasets. This work applies a Random Forest (RF) model and a Long Short-Term Memory (LSTM) model to this dataset to classify drivers. The results show that drivers can be identified with high accuracy, reaching 97.92% for Random Forest and 98.81% for Long Short-Term Memory. These findings indicate that raw CAN data can be suitable for intra-vehicle driver fingerprinting, and the discussion evaluates what dataset characteristics are important for accurate driver fingerprinting, providing insights for future work.