Thi Trang Linh Le - Post-graduate Student, Department IIST FREC, Moscow Institute of Physics and Technology (State University). E-mail: tranglinh2011@gmail.com

Тhearticle consists ofintroduction and three parts: 1. Neural networks in the task of detection of DoS attacks, 2. The experi-mental results and 3.The conclusion.References. 11.In introduction it is emphasized that the purpose of article is demonstration of the fact that in the task of detection of DoS attacks, as well as in the task of the adaptive control, the neural network CMAC is effective alternative of the multilayer neural network (MLP). The article develops results of works about application of the NN CMAC for detection of DoS attacks. Part 1. Neural networks in the task of detection of DoS attacks describes the general approach of application of neural networks for detection of DoS attacks andpeculiarity of application of the NN CMAC and NN MLP for the solution of this task. It is emphasized that as input information for neural networks signs of feature of the network traffic which for the NN CMAC shall be transferred to the digital form owing to properties of this network shall be used. The short description of the main properties of the NN CMAC is provided. Primary part of this article is 2. The experimental results. Here are given training and testing resultsof the NN CMAC and MLP in records of the database of attacks KDD Cup 99 in which repetitions of data beforehand are deleted. For these records values of five features with numbers {3, 4, 5, 23, 30} are selected. For the NN CMAC features are quantized by an even stride so that quantized variables accept values from 1 to the maximum value of . For the MLP features are aligned. The trained neural networks are exposed to testing. The output of a neural network allows to define estimates of probabilities of errors of the first and second kind of system of DoS attacksdetection. The assessment of probability of detection of DoS attacks and share of false alarms for the NN CMAC is equal, respectively, 0,9994 and 0,02, for MLP is equal to 0,9867 and 0,03. In section 3. The conclusion is marked that results of training and testing of two systems of detection of DoS attacks showed that both neural networks are the effective instrument of detection of DoS attacks and in force the of the different nature can be used as a part of neural network expert system of detection of DoS attacks.

 

1. Miller W.T., Glanz F.H., Kraft L.G.CMAC: an associative neural network alternative to backpropagation // Proc. IEEE. 1990. V. 78(10). P. 1561-1567.
2. Tolle H., Ersu E. Neurocontrol. Learning control systems inspired by neural architectures and human problem solving strategies. Springer-Verlag. 1992. 211 p.
3. Cannady J.Next Generation Intrusion Detection: Autonomous Reinforcement Learning of Network Attacks // Proc. of the 23-rd National Information Systems Security Conference, Baltimore, MD, USA, 2000. (http://csrc.nist.gov/nissc/2000/ proceedings/toc.html).
4. Avedjan EH.D., Le T.CH.L.Nejjronnaja set SMAS v zadache obnaruzhenija atak na informacionnye resursy // Informatizacija i svjaz. 2015. № 4. S. 93-98.
5. KDD Cup 1999 Data: http://kdd.ics.uci.edu/databases/kddcup99/kddcup99.html.
6. Albus J.S. A new approach to manipulator control: the cerebellar model articulation controller // ASME Trans., J. Dynamic Systems, Measurement and Control. 1975. V. 97. № 3. P. 220-227.
7. Albus J.S. Data storage in the cerebellar model articulation controller (CMAC) // ASME Trans., J. Dynamic Systems, Measurement and Control. 1975. V. 97. № 3. P. 228-233.
8. Avedjan EH.D. Associativnaja nejjronnaja set SMAS. CH. I. Struktura, obem pamjati, obuchenie i bazisnye funkcii // Informacionnye tekhnologii. 1997. № 5. S. 6-14.
9. Avedjan EH.D. Associativnaja nejjronnaja set SMAS. CHast II. Processy obuchenija, uskorennoe obuchenie, vlijanie pomekh, ustranenie vlijanija pomekh v dvukhslojjnojj seti // Informacionnye tekhnologii. 1997. № 6. S. 16-27.
10. Avedjan EH.D. Algoritmy nastrojjki mnogoslojjnykh nejjronnykh setejj // Avtomatika i telemekhanika. 1995. № 4. S. 106-118.
11. Suratgar A.A., Tavakoli M.B., Hoseinabadi A. Modified Levenberg - Marquardt method for neural networks training // Int. J. of Computer, Electrical, Automation, Control and Information Engineering. 2007. V. 1. № 6. P. 1745-1747.