veoexrmpzt
Dołączył: 21 Lut 2011
Posty: 371
Przeczytał: 0 tematów
Ostrzeżeń: 0/5
Skąd: England
|
 Wysłany: Wto 3:06, 05 Kwi 2011 Temat postu: jimmy choo wedding shoes cxn vqj nlxs cxs |
 |
|
Optimization of RBF neural network based intrusion detection
The entropy of the point x is defined as follows: silk Ei = '(log2S + (1 a) log2 (1 a)); | Ex where i = 1,2, ..., N. Entropy clustering algorithm detailed below. Step one: The data set X = (x,[link widoczny dla zalogowanych], X,, ..., x), were normalized. Step Two: ttlg ~ akx, ux distance D. Step Three: ttg ~ x and x of the similarity S. Step Four: Calculate the entropy of all the sample points E. Step Five: Select the minimum entropy of the sample points as the first m class cluster centers. Step Six: Calculate the step and the similarity of the selected cluster centers o 【big collection of sample points x (set c 【0.7). Step Seven: When the last step in the x, is greater than p = 0.01, in the data set delete set of sample points x and x, when X is less than D, the sample points X in the treated as a separate category,[link widoczny dla zalogowanych], not again as a candidate cluster center, and removed from the sample set. Step Eight: When the x in the data is not empty, go to step five. Entropy clustering algorithm is applied to the RBF neural network to form the E-RBF ~ 0 by the network, we will experiment E-RBF neural network performance. 3 Experimental results and analysis of experimental data from KDD'CUP99 data sets, data sets reflect the four categories of 38 attacks: DOS ① Fund Project: Youth Foundation of Xinyang Normal University (20080205). Author: Guo Xu Show (1976 I) TA,[link widoczny dla zalogowanych], Master research, computer networks and security, artificial intelligence. He Yong (1979) research lecturer, computer network. 6 TECHNOLOGY INFORMATION SCIENCE & TECHNOLOOYINFORMATION class, probing class, R2L class, U2R class. Each contains 41 dimensional feature data, divided into the basic characteristics, content characteristics, the flow characteristics of two seconds, four main flow characteristics. Characteristic features of the data values in a unified and non-numerical feature makes all the numerical data can be encoded neural network processing l3】. After analysis, to play a small role in addition to the data of 11 features: Pr0tOCOLtype, Land, Wrong - fragment, Urgent,[link widoczny dla zalogowanych], Num - failed - logins, ROOt-Shell, Su-attemPted, Num-file-creatiOns, NUm-she118 , Numoutbound_cmds, Is_hosLlogin. Retains the remaining 30-dimensional features. 1149O randomly selected experimental data, pieces of data, covering the four kinds of attacks. One of the 7490 data as training data (including 5000 normal data, 2490 attacks on the data), the rest of the 4000 data as test data (including 2500 normal data,[link widoczny dla zalogowanych], 1500 attacks on the data.) Indicators used to evaluate the detection of two test results. Detection rate = exception has been detected in the number of connections / total number of abnormal connection. False detection rate of abnormal = normal connection connection number of false positives / total number of normal connections. By matlab7.0 simulation test, the detection rate was 92.3% and a false detection rate was 1.62%. The results show that will be introduced into the RBF entropy clustering neural network, can improve the detection rate and reduce the false detection rate. RBF neural networks in intrusion detection is a good effect, in future studies with a very good application and development prospects.
Post został pochwalony 0 razy
|
|
