Аннотации:
© 2018, Institute of Advanced Scientific Research, Inc.. All rights reserved. In this paper, we research the dynamic neural network method for biometric identification of computer users. We analyze the task of detecting the substitution of computer systems users basing the methods of password authentication or authentication using technical devices. To solve this problem, the need to apply biometric authentication methods is actualized. Various methods of users biometric features isolating based on discrete orthogonal transformations are considered. The requirements for choosing biometric identification and authentication methods are formulated: there is no need for additional hardware equipment, the possibility of imperceptible user identification and the analyzed features readability during the workstation use. According on these requirements necessity of users’ recognition based on the mouse moves dynamics is justified. The technique of initial data collecting and their preparation for analysis on the basis of neural network training is described. The neural network model construction use “Deductor” environment. The method of informative features optimal system and neural network architecture searching is developed. We suggested the most efficiency neural network model by the obtained analysis results for computer user’s biometric identification. As the criterion of neural network model optimality the minimal error in user substitution detecting was chosen. The best was a neural network with 6 neurons in the hidden layer, a binary output and 10 input neurons.
