Empirical keystroke analysis in passwords

Rhythmic patterns in passwords are addressed as a kind of biometrics. Experimental results are obtained through two publicly available databases. A preprocessing step (time interval equalization) is applied to both down-down keystroke latency and key hold-down time. Improvements from this preprocessing step are shown through experiments intentionally adapted from papers by the owners of both databases. Afterwards our main experiments are guided by questions Q1: How long does it take for a typist to develop a proper timing signature associated to a new meaningless password? And Q2: How does the number of symbols affect biometric performance? Measurements show that for the password .tie5Roanl typists need many dozens of repetitions to stabilize their typing rhythm. As for question Q2, experimental results show better performance for the shorter password try4-mbs, and that even for longest one studied, .tie5Roanl, there is room for performance improvement.