Skip to Main Content
Over the last six years Pace University has been developing a long-text-input keystroke biométrie system. The system consists of three components: a java applet that collects raw keystroke data over the Internet, a feature extractor, and a pattern classifier. This paper presents two significant system improvements. The first achieves high performance with a closed system of known users and shows how performance changes as the system is opened (diluted) by additional users. The second is the extension of the k-nearest-neighbor classification method to directly derive Receiver Operating Characteristic curves from the classification data. Performance results on 120 participants are presented.