Research On The Data Gravitation Based Classification Method And Network Anormaly Detection Model

By the fast development of computer network in recent years, network intrusion accidents also occur more and more frequently. Traditional network security technologies such as anti-virus and firewall are not very effective in detecting intrusion, so intrusion detection becomes an important research area in network security research system. Among intrusion detection technologies, misuse detection is a technology based on rules. Misuse detection is a kind of low level intrusion detection technology because of its essential limitation. As a high level form of intrusion detection technology, anormaly detection has become the most pop topic in intrusion detection research. The most inportant advantage of anormaly detection is that it can detect unknown intrusions according to known intrusion patterns. Many technologies such as data mining, neural network and support vector machine can be used to build anormaly detection model.This paper introduced universal gravitation and the Law of Gravitation into data classification problem, proposed a novel data classification method called the Data Gravitation based Classification (DGC), and build a high effective anormaly detection model based on DGC. DGC defines the similarity between data points in data space as data gravitation, and there exist data gravitation between any two data points in data space. This standpoint is defferent from the standpoint of traditional data mining technologies, because traditional data mining always treate data points in a local scope. Based on data gravitation, the paper proposed the Law of Data Gravitation, and gave the method of data gravitation computation. By comparing the value of data gravitation, data classification can be performed.Based on the DGC theory, the paper proposed a thinking of weighted features in order to solve the feature selection problem in anormaly detection, and select features by weights in DGC model. A new algorithm called tentative random selection algorithm has been used to optimize the weights of features. This improved DGC method is called...