Joyce and Gupta have described an user authentication system by using keystroke in [JG90]. They are identifying a user by comparing the keystroke latencies of a fixed string, i.e. the password, with the previously stored samples.
Our work differs substantially with Joyce and Gupta's approach because we do not rely on pre-selected strings such as passwords. Further we are continuously learning and refining our model of the user. In addition we have used a finer grained keystroke event model by processing both keypress and keyrelease events. Joyce and Gupta use latencies between keypresses only.