The Research On Intrusion Detection Model On Wavelet Neural Networks

With the abroad applications of the internet, network intrusion and attacks increase rapidly. The issue of computer network security becomes and more severely. IDS (Intrusion Detection System) as a dynamic security.Although the existing IDS can detect the most of attacks, but with the rapid proliferating of network flux the audit data is also increase rapidly, In order to pick up representative characteristic of system mode from a great deal of audit data and construct a intelligent and efficient IDS, this paper adopt a creative IDS based on wavelet neural network improved by rough sets. The method full use rough sets theory as a new mathematical tool to uncertain and vague data analysis has good data mining capability and advantages of neural network, such as learning form itself, association memory and identification unknown intrusion type.Firstly, we have an in-depth study on intrusion detection, rough sets theory and wavelet neural network technologies, and search the feasibility that wavelet neural network and rough sets theory can be applied to anomaly intrusion detection.Secondly, based on academic study we present a network-based intrusion detection system. The emphasis of this thesis is network-based anomaly detection model using wavelet neural network and rough sets theory.Finally, the effectiveness of this model is evaluated in an experiment kddcup99 Intrusion Detection data sets.The computer simulation experiments proved that the IDS based on Wavelet Neural Networks improved by rough sets is better than any other IDS on detection ratio and misreporting rate, when the training data up to 20% of the total data the minimum detection rate and maximum misreporting rate attain to 93.23% and 2.7% respectively, IDS based on rough sets improved a difficult problem on how to choose characters when use neural network in intrusion detection system and "multi-dimensional" problem of WNN. The simulation experiments in this paper proved the identification system based on wavelet neural network has accuracy judgment ability when the quantity and ratio of training network's data is selected properly, In which the attack distinguish rate can up to 97.21% like DoS. So the corresponding taking-precautionary measures of network security may be adopted to aim directly at detected out intrusion action. This is of practically certain value for improving security of the computer communication network.