| With the rapid development of Internet, the problem of Internet security tends tobecome complicated. Traditional security defense technology is hard to meet the needs ofcurrent network safety. As an active security defense technology, intrusion detectiontechnology has been a core technology of the computer safety measures. It providesprevention of insider attack, outsider attack and incorrect operation. With our increasinglydemanding of network communication technology safety, intrusion detection technologyneeds high attention. SVM is a new work machine based on VC theory of statisticallearning theory and structural risk minimization principle. It bases on well-developedtheory, not only has non-linear processing ability, generalization ability and learningperformance but also can solve problems of identifying small sample, non-linear and Highdimension identification. It shows unique advantages and good foreground, so it has beenwidely applied for intrusion detection and model identification.This paper not only has qualitative analysis performance, but also can tackle RS,which includes uncertain knowledge, imperfection data, inconsistent and uncertaininformation, and SVM of the idea of increment learning, they are combined to lead in dataanalysis of intrusion detection, then come up with a new kind of network intrusiondetection method of attributes reduction and incremental SVM. Make full use of RStheory’s advantages in dealing with large data and removing information system(decisiontable), decrease training data of SVM, regard generalization condition KKT as judgment,and make use of the well classification performance of SVM to classify attribute subsetthrough reduct of Rough Set to achieve the purpose of the rapid classification.Put forward intrusion detection method based on SVM-LVQ, combine generalizationcapability of SVM and automatic learning ability of neural network, adopt Eigentransformation and non-Eigen transformation to carry out comparison training andidentification, both method and arithmetic above have been proved by experiment on KDDCUP99intrusion detection DS. Simulation experiment indicates the model has a goodstability and generalization ability, keep better detection accuracy, as well as improve thespeed of training and detection, which reflects more availability. |
