The ATVS-Keystroke database is a dataset captured for performance evaluation of Keystroke Dynamics recognition systems (see [BTAS15] for all the details). The database comprises 63 users with 12 genuine access and 12 impostor access for each user for a total number of samples equal to 7680 (63 users x 24 access x 5 data). There are people from two different nationalities with 60% of males and 40% females. The acquisition was made in two sessions according a semi-supervised protocol:

1. First session: the users were asked to introduce their personal data in the platform. This process was repeated six times.

2. Second session: after at least 24 hours, the users were asked to introduce once again their personal data in the platform. The process was repeated six times. In addition, in this second session, each user acted as an impostor trying to spoof the system with the personal data of another user. The personal data of three other users was showed to each of the impostor and they introduced them four times for a total number of impostor access of twelve per user.

The information provided by the users includes sensitive data and therefore, it has been post-processed to remove all the personal information (the characters pressed) and to maintain the privacy of the users enrolled in the database. The keystroke dynamic patterns were recorded using a key-logger (programmed in Java). The key-logger detects two different types of events: press and release. The timestamps for each of the detected events were recorded in milliseconds.

For further information on the database we refer the reader to (the following article is publicly available in the publications section of the ATVS group webpage .)

Please remember to reference [BTAS15] on any work made public, whatever the form, based directly or indirectly on any part of the ATVS-Keystroke DB.