Abstract:

As the support vector machine (SVM) classification approach has a good generalization performance in the cases of small  number and non-linear samples, it is widely used in network intrusion detection fields. In order to resolve the offset  phenomenon of separating a hyperplane caused by imbalanced data, Riemannian geometry inherent in a nuclear function is  regarded as an important basis and a pseudoconsistency transformation function is also introduced, both of which are  used to modify the kernel function and improve the generalization ability of SVM classification. On this basis, an  intrusion detection system based on modified kernel function SVM is established, and a detailed description of the  overall structure of the system and operating mechanism is made. Finally, simulation experiment shows that this system  can achieve a more accurate detection rate and improve the SVM’s classification offset phenomenon caused by imbalanced  data sets.