Keystroke Dynamics Based Biometric Identification

Abstract

Biometrics based keystroke dynamics aim to perform user identification and authentication based on users' typing behaviour on digital devices. In this study, keystroke timing and regional distributions extracted from free-text are utilized to perform user identification. In order to obtain the highest representative set of attributes, attributes based on directional graph, hold time and keyboard distance have been extracted and used in different configurations. In order to process the generated feature sets more effectively, unlike the existing studies, a multilayer artificial neural network model with attention mechanism was used and 0.13% FAR and 2.5% FRR results were obtained.