| With the rapid development of network technology, the issue of networksecurity becomes more and more serious, and it has been difficult to deal withcomplicated network attacks for the traditional intrusion detection technology.Moreover, the self-adaption, self-organization and distribution are thecharacteristic of biology immune system, which is also the traditional intrusiondetection technology wanted. Therefore, intrusion technology base on theimmunity having attracted more and more concern of experts and scholars.Intrusion detection systems often have to face a lot of large-scale data,intrusion detection base on the immune system is no exception. Intrusiondetection system based on immunity detects the attack by the detector,performance of detector influence the detection results directly. But thereal-valued detectors have poor performance in detecting high dimensional data.To this issue, this paper introduced information gain theory used in textcategorization to intrusion intrusion detection base on the immunity.Firstly, through the analysis of curse of dimensionality in high-dimensionalspace, we find that the reason of the real-value detector in a high-dimensionalspace is the small detection range and uneven distribution, in this paper use theattribute selection base on the information gain to control the detector/self setdimension, ensure the shapspace in a reasonable dimension.Secondly, for the problem of poor performance of the real-value detector indetecting high dimensional data, this paper proposes a detector generationalgorithm based on information gain, this algorithm through select attribute withhigh information gain to restruct a new feature space, and generate detector innew feature space, which make detectors in the low dimensional space could detect the anomaly data in high-dimensional.At last, we take the off-line dataset Kddcup99to prove the feasibility ofdetector generation algorithm based on information gain. |
PDF Full Text Request