Keystroke Biometric: Data Capture Resolution Accuracy
Michael Lam, Upen Patel, Mark Schepp, Teresa Brooks Taylor, Robert S. Zack · 2010
This study extends work on the Keystroke Biometric Authentication System developed in Pace University’s Seidenberg School of CSIS. This system can identify with a high degree of accuracy the typing characteristics that are unique to an individual. The system consists of three components: a Java applet which collects raw keystroke data over the internet, a feature extractor, and a pattern classifier. The aim of this study is to examine the available methods of capturing keystroke data, determine the accuracy of the data captured by the existing Java applet, and compare the accuracy of our system with the reported accuracy of other keystroke systems. In more than half of the 36 machines tested we found that the millisecond digit of the clock was frozen or inaccurate, resulting in a recording accuracy of centiseconds rather than milliseconds in those machines. Another finding was that the dwell times (key press durations) were roughly normally distributed and the flight times (from the press one one key to the press of the next) were roughly log normal distributed.