Font Size: a A A

Research Of IDS Based On Hybrid Neural Network Algorithm

Posted on:2011-01-28Degree:MasterType:Thesis
Country:ChinaCandidate:J H DaiFull Text:PDF
GTID:2248330395458325Subject:Computer application technology
Abstract/Summary:PDF Full Text Request
During the past two decades, the explosive development of Internet brings infinite opportunities to societies, economics and cultures. In the meanwhile, it also brings rigorous challenge to information security. People use anti-virus, firewalls, intrusion detection technologies and other technical methods to ensure the security of network information. With the development of network technology, the intrusion detection technology has become a critical component of network security architecture.In this thesis, several kinds of artificial neural network algorithms are analyzed methodically. Those are the most popular algorithms in a morden intrusion detection engine. Through the analysis, we can contrast their advantage and shortcoming. Base on this research, we proposed a hybrid neural network model which includes a SOM module and a BPN module. Using the feature of self-organizing, our intrusion dectection system can classify intrusion and decrease rate of false positive alarm. Result of this SOM module would be the input vector of back propagation algorithm, using the advantage of BPN to cluster the category of attacks; we can control the rate of false negative alarm effectively.In order to speed up the proposed model further, we introduce a PCA algorithm to reduce the dimensions of SOM nework’s output vectors. After the dimensions reduction, these lower dimensions’vectors still maintain the orginal properties. Use this new version vectors as the input of BP module can reduce the need of calculation dramatically which can get a better perform in real-time applications.During the simulation and testing, Firstly, we did a contrast evaluation between our hybrid neural network algorithm and classic ANN algorithms. The result show us not only this hybrid intrusion detection system perform a more useable false alarm rate, but also some advantages of hybrid structure improve the visualization function of IDS. Secondly, we contrast the modified PCA algorithm and our orginal hybrid model; result shows us that this PCA algorithm modification can help IDS improve its real-time performance without sacrifice intolerance accuracy.
Keywords/Search Tags:Intrusion detection system, Hybrid neural network, BP neural network, SOMneural network, Principal Components Analysis
PDF Full Text Request
Related items