| Summary: | 碩士 === 國立雲林科技大學 === 資訊管理系碩士班 === 95 === The identity authentication is a very important issue on computer systems. The most usual mechanism used to recognize a legitimate user at present is the Login Password authentication. This authentication is fragile if someone else gets to know the password. The Biometrical mechanism is obviously the strongest way to authenticate people; however they usually require expensive hardware to support the dedicated function. Keystroke dynamics is another biometrical measure that has been explored lately as a mean to easily support user identification. Notwithstanding, it is not a method as robust and effective as others that involve costly equipment and software, it is not difficult to obtain the information from users. Thus the keystroke authentication has become a very useful way to enhance password authentication on computer systems. Keystroke biometrics is presently one of the less-studied biometrics and researchers tend to collect their own data, so few studies have compared recognition techniques on a common database. Therefore this paper attempts to discover the most appropriate configuration for implementation of keystroke authentication mechanism. In this paper we implemented three familiar methods for keystroke identification including statistical method, back-propagation neural network classifier, and decision tree classifier. ANOVA-test for statistical testing was used to distinguish whether there is a significant difference between the configurations of these methods. Scheffe post-hoc test was used as well to help us to indicate which configuration achieved the highest accuracy rate. Furthermore, we proposed a novel method for keystroke authentication as well to increase the accuracy rate. The results suggest that excellent accuracy rate is obtained from our method.
|
|---|
