Network Intrusion Detection Based on GMKL Algorithm.
Yuxiang Li,Haiming Wang,Hongkui Yu,Changquan Ren,Qingjia Geng
DOI: https://doi.org/10.4304/jnw.8.6.1322-1328
2013-01-01
Abstract:According to the 31th statistical reports of China Internet network information center (CNNIC), by the end of December 2012, the number of Chinese netizens has reached 564 million, and the scale of mobile Internet users also reached 420 million. But when the network brings great convenience to people's life, it also brings huge threat in the life of people. So through collecting and analyzing the information in the computer system or network we can detect any possible behaviors that can damage the availability, integrity and confidentiality of the computer resource, and make timely treatment to these behaviors which have important research significance to improve the operation environment of network and network service. At present, the Neural Network, Support Vector machine (SVM) and Hidden Markov Model, Fuzzy inference and Genetic Algorithms are introduced into the research of network intrusion detection, trying to build a healthy and secure network operation environment. But most of these algorithms are based on the total sample and it also hypothesizes that the number of the sample is infinity. But in the field of network intrusion the collected data often cannot meet the above requirements. It often shows high latitudes, variability and small sample characteristics. For these data using traditional machine learning methods are hard to get ideal results. In view of this, this paper proposed a Generalized Multi-Kernel Learning method to applied to network intrusion detection. The Generalized Multi-Kernel Learning method can be well applied to large scale sample data, dimension complex, containing a large number of heterogeneous information and so on. The experimental results show that applying GMKL to network attack detection has high classification precision and low abnormal practical precision.