Support Vector Machine And Its Application In Intrusion Detection

Support Vector Machine(SVM) is a new method of machine learning. It is based on the statistical learning theory and performs excellently in classification, regression, time prediction problems and so on. Now, SVM has been applied into many application fields such as text classification, handwritten digit recognition, face recognition, intrusion detection. The paper studies some algorithms about SVM in the application of intrusion detection problem.Intrusion detection techniques must be adapted to the high-speed and distributed network environments in the future. In the above network situations SVM needs to learn much training data and detect intrusions on time. However, there is another problem that the abnormal class's distribution is disperse and is less than the normal samples. The paper analyzes the rapid PSVM algorithm and the influence of unbalanced class to the classification effectiveness, uses the weighted method to improve the linear and nonlinear PSVM. Experiments on the standard dataset show that the training time is less and the precision of classification is better.Feature selection and extraction are important bases of applications of machine learning algorithms. The paper uses kernel principal component analysis to extract features from the intrusion detection training samples. The method extracts features and reduces the dimensions very effectively. In addition, we make use of RSVM method into nonlinear proximal SVM. It can reduce the computation requirements of the kernel matrix. The combination of the above two methods improve the training speed and classification effect.Incremental learning has been a research focus in recent years, its advantage is that the learning process can discard useless samples automatically, reduce the training set and save storage costs. Though classical or standard SVM algorithm does not have incremental learning ability, its theoretical system of support vector concept is great significant to incremental learning construction. The paper discusses the use of the KKT conditions and changes of support vectors, non-support vectors during the incremental learning on the basis of the relations of support vectors, non-support vectors with classification hyper-plane and separate hyper-planes. We also give proof of some relevant theories and compare effectiveness of several incremental learning methods. In the end, we suggest an incremental learning algorithm based on Lagrange SVM which can solve classification problem in the unbalanced situation. Experiments on the standard dataset show that it has better precisions to the negative class and the time is comparable to the traditional incremental learning method.