The entire dataset is available now. Download and sign the EULA.
The Tutorial of the Ongoing Evaluation is available (download the tutorial or visit the video tutorial)
The Ongoing Evaluation is available (see the Ongoing Evaluation site for more information)
The Database
The database consists of keystroke sequences from 300 subjects acquired in four different sessions distributed in a four month time span. Thus, three different levels of temporal variability are taken into account: (1) within the same session (the samples are not acquired consecutively), (2) within weeks (between two consecutive sessions), and (3) within months (between non-consecutive sessions).
Each session comprises 4 case-insensitive repetitions of the subject’s name and surname typed in a natural and continuous manner (case insensitive). Mistakes (and backspace) are not permitted. The time (in milliseconds) elapsed between consecutive key events (press and release) is provided as keystroke dynamics sequence. The main statistics of the dataset proposed for the competition are summarized in Table 1.
Table 1. Summary of the main statistics of the database proposed for the competition
There are two types of sequences:
Genuine sequences: typed by the owner of the password.
Impostor sequences: typed by other users (different to the owner) who try to spoof the system.
The number of genuine and impostor samples per user varies (this information remains sequestered). Each file sample is named as TXXX_YY.txt, where T is the set (D=Development, T=Test), XXX is the number of user and YY the number of sample. The files contain the following information:
23 %Number of keys
0x2aD 0 %0x2a=scancode (see examples in the development set), D=type of event(D= press, U= release), 0=time between two consecutive key events (the first is always the reference zero).
0x24D 265
0x24U 109
0x2aU 0
…
The time sequence is given in form of time (milliseconds) elapsed between events. In this example, the difference between the pressure of the key 0x2a and the key 0x24 is 265 milliseconds, the difference between the pressure of the key 0x24 and the release of 0x24 is 109 milliseconds,...
The Evaluation
The first 4 samples are used to model the user behavior (files TXXX_01.txt, TXXX_02.txt, TXXX_03.txt, TXXX_04.txt are genuines) and the remaining 20 samples (including genuine and impostors) are used to calculate the performance of the systems. Therefore, the evaluation comprises 20x300=6000 comparisons. See all the details in:
Aythami Morales, Julian Fierrez, Ruben Tolosana, Javier Ortega-Garcia, Javier Galbally, Marta Gomez-Barrero, Andre Anjos, Sebastien Marcel, "Keystroke Biometrics OnGoing Competition", IEEE Access, Vol. 4, pp. 7736-7746, November 2016.
Download the entire dataset for the Offline Evaluation of your algorithms (include a Matlab script example): You must download the EULA (see at the bottom of the page), sign and send to kboc2016competition@gmail.com with the Subject: "KBOC EULA". They will receive the link to download the entire KBOC dataset (available soon).
Ongoing Participation (more details at Ongoing Evaluation site): https://www.beat-eu.org/platform/experiments/robertodaza/robertodaza/competition_kboc16/2/Kboc16_Competition_Baseline_Modified_Scaled_Manhattan_Distance/
Tutorial of the Ongoing Evaluation: available now, download the tutorial or visit the video tutorial