Intrusion Detection Based On Least Squares Support Vector Machine

Along with the rapid development of Computer network, the problem of network security is emerging increasingly today. Although as a kind of active supply of traditional safety mechanism, intrusion Detection acts as the effective complement to traditional protection techniques, traditional intrusion Detection model reveals more and more limitation in the face of increscent network flux,update network equipment new attacks。We have do some research about intrusion Detection In this dissertation, research about intrusion Detection In this dissertation.Least squares support vector machines is a famous kind of extension of support vector machine (SVM). Least squares support vector machines have not robustness, when converting the inequality constraints into equality constraints in Vapnik equation and introducing of the concept of relaxation factor, although it can avoid solving problems of quadratic Programming and greatly reduce the amount of calculation, the sparse of SVM is lost and succeed noise-sensitive.In order to overcome the drawbacks of not sparse and noise-sensitive, we proposed to use of FSALS-SVM which is a rapid Sparse method. It iteratively builds the decision function by adding one basis functionfrom a kernel-based dictionary at one time, the process is terminated by using a flexible and stable spsion insensitive stopping criterion. First, in order to guarantee dates can reflect real condition of the model that we have proposed, a flow that is using means of HVDM distance metric of heterogeneous datasets to preprocess the feature data of network is proposed, then,based on traditional CIDF Frame-model, we have taken into account the components of the data acquisition, capability, response and soon. At last we Propose LS-SVM intrusion detection model. The simulation experiment results show that LS-SVM for intrusion detection is feasible and effective and it has certain advantages compared with SVM.The size of train set is also a factor to estimate classifiers effectively. When the size of train set is too small, it can not truly reflect classifiers performance; when the size of train set is too big, there is a great amount of computation.So it is necessary to discuss the size of train set. Similarly, some data feature of heterogeneous data integration has a great influence on classifiers, or even to determine the result. some has no influence on classifiers.Taking into account data feature of heterogeneous data integration has different influence on classifiers, we presented a weighted processing method that we put Weighted Function which is designed by our self into Pretreatment process flow during the data processing Procedure. The simulation experiment results reveal that The technique of Pre-extracting we proposed is useful to get a suitable size of train set for classifiers.With the increasing of the size of train set, predictive error ratio will be reduced to a certain extent which is our purpose to select the size of train set, it also illustrates the detection precision has improved to a certain degree after the treatment of Weighted Function.