Statistical Learning Theory (SLT) is a new statistical theory framework established from finite samples. SLT provides a powerful theory fundament to salve machine learning problem with small samples. Support Vector Machine(SVM) is a novel powerful machine learning method based on SLT.SVM solves practical problems such as small samples, nonlinearity, over learning ,high dimension and local minima, which exist in most of learning methods, and has high generalization.As an embranchment of the SVM, LSSVM inherit some research production of SVM in theory and application. Relative to the SVM, the characteristic of LSSVM could be summed as explaining the linearity equations, faster explaining speed, needing less resource, the answer fulfilling the minimum.This thesis firstly expatiates the background and signification, introducing the model and the sort of the intrusion detection, comparing the advantage and disadvantage of each kind of detection technology which applied in intrusion detection, then introduce the relative theory of the SLT and SVM and LSSVM, put forward a network intrusion detection model based on the LSSVM, and discuss thoroughly on the function and mechanism and realization of each component. By means of HVDM distance metric of heterogeneous datasets, the feature data of network are preprocessed. Aiming at the disadvantage of losing the robust of the LSSVM, a weighted least squares support vector machine is proposed in this thesis. Since the empirical risk is calculated via quadratic function, LSSVM loses sparseness of SVM and this leads to the decrease of calculation efficiency when classifying. To spare LSSVM, the principal component analysis method was used to extract feature , clear up the irrelevance and select import examples of training sample as support vector (SV),and the information of non SV examples was transformed to SV, so new sparse algorithm was proposedâ€”PCA-LSSVM. In the simulative experiment, RBF was selected to map the training data from low dimension to high dimension, so the data could be divided in high dimension, Three Step Search algorithm was used to choose the parameter and SMO was used to train the dataset. The result shows that the sparse LSSVM classifier keeps the classify ability of the SVM, and the sparse rate was higher, the SV count was less than standard support vector, enhance the classify efficiency and the real-time of the LSSVM obviously. |