Intelligent Intrusion Detection Model Based On Information Gain Bayesian Network

With the rapid development of the Internet, network security is becoming more and more important. The technology of Intrusion Detection is presented under the situation that traditional security strategies are incapable of satisfying ever increasingly rigorous demand. Intrusion Detection is becoming a hot topic as an important component of Network security.Most of current models of intrusion detection have some defections such as lack of adaptability and intelligence. On the basis of current intelligent intrusion detection models have some advantages and some disadvantages. The dissertation proposes the model of Naive Bayesian network intelligent intrusion detection and implements it, and introduces the theories of naive Bayesian network. Basing on the model of Naive Bayesian network intelligent intrusion detection mainly finishes three researches as following:1. The dissertation proposes intelligent intrusion detection model based on information gain Naive Bayesian. We select useful attribute as classifiable feature by using the information gain to delete the redundancy attribute in order that the model has the ability to select automatically. Through deleting redundancy attribute, we can reduce the complication of Naive Bayesian network algorithm and improve detection rate.2. In order to make model quickly adapt new attacks and improve model's self-adaptability, the dissertation designs and implements the model of increment naive Bayesian network. When new attack appears we can base on the model originally trained. Through increment Naive Bayesian network only studynew added instances to modify the parameter of NB .It can reduce training time and the model has the ability of self-learning and can be adaptable to various environments.3. The foregoing models are based on naive Bayesian network. There is a powerful hypothesis that all features are independent with each other. This independence assumption does not satisfy the reality. Hence, we make the model adapt the general situation. By relaxing independence assumptions, using the algorithm studies the structure of Bayesian network. The trained BN can reflect the true situation and make the model apply to more situations.The intelligence of foregoing models lies in the factors of the model can be adjusted by the comparison of similarity between the data. Thus, the system has the ability of self-learning and can be adaptable to various environments. The research focuses on what we said. The experiments show that the foregoing models own better performance.