[114] Integral evaluation of the dangerous behaviour probability for computer network users via parameters of their activity
L S Kuravsky1, G A Yuryev1, M A Chervonenkis2, A A Konstantinovsky1, A A Shevchenko1, S S Isakov1 and D P Sincha2 1Moscow State University of Psychology and Education, Computer Science Faculty, Moscow, Russia 2Pawlin Technologies Company, Moscow, Russia. Email: l.s.kuravsky@gmail.com
Under consideration is an integral evaluation of the dangerous behaviour probability for computer network users, which is based on three independent techniques, viz: target statistics generated by the self-organising feature maps (with the following statistical hypothesis testing), maximum likelihood estimations calculated with the aid of the Markov chains and assessments obtained by the long short-term memory (LSTM) recurrent neural networks. Probabilistic indicators are derived for three different assessment responsibility levels. The results obtained are illustrated with examples of practical work. A comparison of the techniques in use is presented.
Keywords: computer network threats, user activity, self-organising feature maps, Markov chains, LSTM recurrent neural networks.
