Multivariate Network Intrusion Detection Methods Based on Machine Learning

Recent years, it is a spurt of progress in the Internet, network security problems such as cyber-attacks have emerged in an endless stream. The activity of an attacker to illegally use and destroy resources on the Internet called Intrusion. To keep the security of network, it is necessary to detect before the intrusion has caused damages. Intrusion detection is a technology that identifies the use of resources and gives early warning of abnormalities, which is an effective complement to firewalls. Early researches were mostly based on statistical and rule-based approaches, using Pattern Recognition. With the development of machine learning, there are two methods, the first methods use conventional machine learning algorithms such as decision trees, KNN and SVM, and the second methods are based on Neural Network models such as CNN. To further improve the feature extraction ability of the model, this paper combines the advantages of traditional machine learning and neural network, applies them to intrusion detection and 2 models with better training effects are finally obtained. We also use different network architectures in the second scheme such as ResNet and DenseNet to verify the validity of the results, using Dropout strategy to improve the generalization ability and reduce the over-fit phenomena.