The Research Of Intrusion Detection Based On Artificial Neural Network

With the daily growing security problems on the Internet, the Network Security Issue has attracted much more attention. The mostly used techniques to protect network facilities are Firewall and Intrusion Detection System. All of the current Intrusion Detection Systems are far from mature. Some of particular products have problems with mass false reports and missing reports of intrusions. Any more, the real-time detection can not meet requirements. In order to solve the real-time problems, the tri-tier feed forward Radial Basis Function (RBF) neural network is used in intrusion detection technology, which is better to enhance the real-time efficiency of Intrusion Detection in a certain extent and debase the rate of false reports and missing reports.The Intrusion Detection System based on the neural network has capabilities of self-learning, memory and fuzzy computing; it can detect not only existing attack mode, but also unknown ones. RBF which is one of the neural network algorithms is widely used in data classification, pattern recognition and many other areas. Compared with the common BP neural network arithmetic, RBF neural network can arbitrary approximate nonlinear functions and solve the inherent laws of the function itself by rule and line. If used in intrusion detection system, its faster convergence can solve the real-time approximating and applying problems of unwonted model.The entire Intrusion Detection System is huge, it typically includes network data packet capturing module, network protocol analysis modules, memory modules, the responding module, intrusion detection module, rules analysis module and interface management module. Among those the rules analysis module is the key one which is analyzed and researched in the paper. The data sets KDD CUP 1999 (SR KDD99) established by the United States Lincoln Laboratory is selected as training and testing data. Using RBF network classifier, the extractive concentrated 41 eigenvalues of the data are abstracted and then set into group to sort according to function, the eigenvalues of weak influence to intrusion detection are filtered and the available ones are used as the RBF neural network input vector to establish the invasion analysis rules through training and testing.Using matlab7.0 RBF neural network toolbox for simulation, the experimental data are obtained. After analyzing and organizing the statistical experimental data, it's proved that intrusion detection based on the RBF neural network have the characteristics of real-time, low rate of false and missing reports.