The Kidmose CANid Dataset (KCID)

Kidmose CANid Dataset (KCID)

The Kidmose CANid Dataset (KCID) contains CAN bus data collected by Brooke and Andreas Kidmose from 16 different drivers across 4 different vehicles. This dataset is designed to support driver identification and authentication research.
The term "CANid" reflects the dataset's dual purpose: data collected from the CAN bus for driver identification research.
VEHICLESThe dataset includes data from four different vehicles across various manufacturers and model years:
2011 Chevrolet Traverse - 5-door full-size SUV crossover, AWD, 8 drivers (8 unique drivers in single-driver traces; 1 additional driver in a mixed trace)2017 Ford Focus - 5-door compact station wagon, FWD, 4 drivers2017 Subaru Forester - 5-door compact SUV crossover, AWD, 6 drivers (6 unique drivers in single-driver traces; 3 additional drivers in mixed traces)2022 Honda CR-V Touring - 5-door compact SUV crossover, AWD, 1 driverNote: The number of drivers includes volunteer drivers whose data was captured in single-driver traces, where we know who was driving at all times. We exclude volunteer drivers whose data is only available in mixed traces because we do not know when each specific driver was actually operating the vehicle.
DRIVERSThe dataset includes 16 drivers across different demographic categories:
Male Drivers:
Under 30 years: 4 drivers ("male-under30-1" through "male-under30-4")30-55 years: 4 drivers ("male-30-55-1" through "male-30-55-4")Over 55 years: 3 drivers ("male-over55-1" through "male-over55-3")Female Drivers:
All ages: 5 drivers ("female-all-ages-1" through "female-all-ages-5")Driver Directory Structure: Driver identifiers are used as directory/folder names. Within each directory, you will find traces collected from that particular driver, with additional information (location, data collection method, etc.) specified in the filename.
Note: We use "unknown driver(s)" in directory names when we know that one or more volunteer drivers was operating the vehicle, but we cannot identify who was driving or when. We used a standalone data logger for some data collection sessions. If we failed to download the data and clear the logger's memory before switching drivers, this resulted in mixed traces and, occasionally, "unknown driver(s)" entries. Unfortunately, some of our volunteer drivers were short-term visitors, so we did not have the opportunity to redo their traces as single-driver traces.
LOCATIONSData collection took place across multiple locations:
DK - DenmarkUSA - United States of AmericaFL - FloridaNE - NebraskaNE-to-FL - Trip from Nebraska to FloridaTN - TennesseeTN-to-NE - Trip from Tennessee to NebraskaLocation codes appear in filenames (e.g., USA-FL-CANEdge-00000001.mf4 indicates data collected in Florida, USA).
DATA COLLECTION METHODSThree different data collection methods were employed:
CANEdge - CSS Electronics CANEdge2: Standalone data logger that connects to the OBD-II port and logs to an SD cardKorlan - Korlan USB2CAN: CAN-to-USB cable connecting the vehicle's OBD-II port to a laptopKvaser - Kvaser Hybrid CAN-LIN: CAN-to-USB cable connecting the vehicle's OBD-II port to a laptopThe data collection method is indicated in filenames (e.g., USA-FL-CANEdge-00000001.mf4).
FILE TYPESThe dataset provides data in three formats to support different use cases:
.mf4 (MDF4) Format: Measurement Data Format version 4 (MDF4)
Binary format standardized by the Association for Standardization of Automation (ASAM)Advantages: Compact size, popular with automotive/CAN toolsUse case: Native format from CSS Electronics CANEdge2Reference: https://www.csselectronics.com/pages/mf4-mdf4-measurement-data-format.log Format: Text-based log format
Compatibility: Linux SocketCAN can-utilsAdvantages: Compatibility with SocketCAN can-utils; if a .log file is replayed, then data can be captured and monitored using Python's python-can libraryReferences: https://github.com/linux-can/can-utils, https://packages.debian.org/sid/can-utils, https://python-can.readthedocs.io/en/stable/.csv Format: Text-based comma-separated values (CSV) format
Advantages: Easy to load with Python using the pandas library; easy to use with Python-based machine learning frameworks (e.g., scikit-learn, Keras, TensorFlow, PyTorch)Usage: Load with Python pandas: pd.read_csv()Reference: https://pandas.pydata.org/SPECIALIZED EXPERIMENTSThe KCID Dataset includes five specialized experiments:
Fixed Routes Experiment
Vehicles: 2011 Chevrolet Traverse, 2017 Subaru ForesterDrivers: male-30-55-3, male-30-55-4, male-over55-1, female-all-ages-1, female-all-ages-2, female-all-ages-5Location: Florida, USA (specific routes)Data Collection Methods: CSS Electronics CANEdge2, Kvaser Hybrid CAN-LINPurpose: Capture CAN traces for specific, mappable routes; eliminate route-based variations in driver authentication data (e.g., low-speed local routes vs. high-speed long-distance routes)OBD Requests and Responses Experiment
Vehicle: 2011 Chevrolet TraverseDriver: female-all-ages-5Location: Florida, USAData Collection Method: CSS Electronics CANEdge2Purpose: Capture OBD requests and responses Arbitration IDs: Requests: 0x7DF, Responses: 0x7E8Tire Pressure Experiment
Vehicle: 2011 Chevrolet TraverseDriver: female-all-ages-5Location: Florida, USAData Collection Method: Kvaser Hybrid CAN-LINPurpose: Capture normal and low tire pressure scenariosApplications: Detect tire pressure issues via CAN bus analysis; develop predictive maintenance strategiesDriving Modes and Features Experiment
Vehicle: 2017 Ford FocusDriver: male-30-55-1Location: DenmarkData Collection Method: Korlan USB2CANPurpose: Capture different driving (and non-driving) modes and featuresExamples: gear (park, reverse, neutral, drive, sport); headlights on/offStationary Vehicles Experiment
Vehicles: 2024 Chevrolet Malibu, 2025 Toyota CorollaDriver: N/A (vehicles remained stationary)Location: Florida, USAData Collection Method: Kvaser Hybrid CAN-LINPurpose: Capture CAN bus traffic from very new, very modern vehicles; identify differences between an older vehicle's CAN bus (e.g., 2011 Chevrolet Traverse), and a newer vehicle's CAN bus (e.g., 2024 Chevrolet Malibu)ADDITIONAL DOCUMENTATIONEach "specialized experiment" directory contains a detailed README.md file with specific information about the experiment and the data collected.
RESEARCH APPLICATIONSThis dataset supports various research areas:
Driver authentication, driver fingerprintingBehavioral biometrics in the automotive domainVehicle diagnostics and predictive maintenanceMachine learning in the automotive domainCAN bus analysis and reverse engineeringCITATIONIf you use the Kidmose CANid Dataset in your research, please cite appropriately. Citation information will be updated when our paper is published in a peer-reviewed venue.
Article Citation:
APA Style: Kidmose, B. E., Kidmose, A. B., and Zou, C. C. (2025). A critical roadmap to driver authentication via CAN bus: Dataset review, introduction of the Kidmose CANid Dataset (KCID), and proof of concept. *arXiv*. https://arxiv.org/pdf/2510.25856
MLA Style: Kidmose, Brooke Elizabeth, Andreas Brasen Kidmose, and Cliff C. Zou. "A Critical Roadmap to Driver Authentication via CAN Bus: Dataset Review, Introduction of the Kidmose CANid Dataset (KCID), and Proof of Concept." *arXiv*, 2025. doi:10.48550/arXiv.2510.25856
Chicago Style: Kidmose, Brooke Elizabeth, Andreas Brasen Kidmose, and Cliff C. Zou. "A Critical Roadmap to Driver Authentication via CAN Bus: Dataset Review, Introduction of the Kidmose CANid Dataset (KCID), and Proof of Concept." *arXiv* (2025). doi:10.48550/arXiv.2510.25856
Dataset Citation:
APA Style: Kidmose, B. E. and Kidmose, A. B. (2025). Kidmose CANid Dataset (KCID) v1. \[Data set\]. Technical University of Denmark. https://doi.org/10.11583/DTU.30483005.v1
MLA Style: Kidmose, Brooke Elizabeth, and Andreas Brasen Kidmose. "Kidmose CANid Dataset (KCID) v1." Technical University of Denmark, 30 Oct. 2025. Web. {Date accessed in dd mmm yyyy format}. doi:10.11583/DTU.30483005.v1
Chicago Style: Kidmose, Brooke Elizabeth, and Andreas Brasen Kidmose. 2025. "Kidmose CANid Dataset (KCID) v1." Technical University of Denmark. doi:10.11583/DTU.30483005.v1