Design And Implementation Of Malicious Domain Name Detection Model Based On Generative Adversarial Network

A botnet is an active network consisting of a master and a large number of controlled terminals,which can be used by attackers to launch DDOS(Distributed Denial of Service),malware download and network mining attacks on target systems.Under normal circumstances,botnets will use Domain-Flux technology to avoid detection,and the domain name generation algorithm(DGA)is the specific implementation of this technology.Attackers can use domain name generation algorithms to quickly generate a large number of DGA domain names in a short period of time.Traditional methods use machine learning algorithms and deep neural networks to detect DGA domain names.However,the final detection effect of these models is not good.In view of this,this thesis proposes a malicious domain name detection model based on Generative Adversarial Network(GAN).The main research contents of this thesis include:(1)Aiming at the problem that traditional methods have low accuracy in identifying malicious domain names,a discriminative network based on conditional classification is designed,which enables the detection model to accurately identify malicious domain names.(2)Aiming at the problem that traditional methods cannot predict malicious domain names,a domain name generation network based on deep feature mining is designed.This network can enable the detection model to simulate the generation of malicious domain names.(3)Research on the application of a malicious domain name detection model based on Generative Adversarial Network.In this thesis,the structural analysis and theoretical derivation of the proposed model are carried out,while statistical comparison experiments are designed to test the domain name prediction function of the generator module,and various classification experiments are also designed to test the domain name identification performance of the discriminator module.The experimental results show that the model proposed in this thesis improves 5.38%over the traditional detection method in terms of classification and detection accuracy of domain names,while the generated samples output from this model are also reasonable and effective.