Biometrics, the computer-based identification of an individual, is becoming increasingly essential due to the increasing demand for high-security systems. Accurate and effective keystroke biometric technology can help provide a major boost to the security of electronic commerce, and it can help curb identify theft. This dissertation focused on keystroke biometric applications operating over the Internet and requiring a text input of several hundred characters. These applications can help identify a perpetrator of inappropriate or fraudulent Internet activity, such as the sending of obscene or inappropriate email. The system developed contained three components. A Java applet collected the raw keystroke data over the Internet, a feature extractor calculated appropriate long-text measurements from the raw data, and a pattern classification system was trained to make the appropriate application-dependent decisions. Experimental results on the keystroke biometric system with 30 users under ideal conditions yielded recognition accuracy of 99.3% showing that our system is highly effective (our "System 3"). The ideal conditions were that a user types a predefined text (copy task) of approximately 600 characters and uses the same keyboard type for both training and testing. Classification accuracy as a function of input text length was also obtained to determine the text length needed to obtain the accuracy required for a particular application.
