Research On Intrusion Detection Model Based On Neural Network Learning From Honey-Pot

With the advancement of science and technology and the development of computer network techniques, more and more people begin to make use of network and enjoy the convenience which is brought by network. With expanding of the scale of network, the security issue has become the core issue that can not be ignored in the internet development.Although the traditional secularity technology, such as authentication and authorization, access control, encrypting information, VPN, Firewall and so on, strengthen the security of sensitive data in computer system in some extent, they can't prevent the user who has been accredited from abusing the computer and making the data filched illegally. So the the technology of intrusion detection appeared. It makes up for security protection measures about firewall and data encryption. It can identify malicious intention and act to the computer and network resources, and make an instant response.Aiming at the problem of high rate of false negatives and false positives of IDS, this thesis presents a kind of intrusion detection model of based on neural network by learning from honey pot. The model applies the neural Network to intrusion detection system to make the system have adapting-oneself and teaching-oneself captivity. The learning algorithm based on genetic ANN is improved. It realizes parallel research and optimizes the learning algorithm. The cooperation of Honey pot and intrusion detection system can decrease the false alarm and false negative rates. Because the model is bridled on the normal and abnormal specimens it decreases greatly the false alarm and false negative rates. A new intrusion detection model is designed in this thesis and the process of realization of this model is introduced in detail. And finally the data is analyzed through simulation experiment. The compare between the traditional model of intrusion detection and this model is made at the end of the thesis. The experiment result shows the model has a good effect, rapid learning speed. The false alarm and false negative rates are very low.