Research On Botnet Detection Technology Based On Deep Learning

In recent years,botnet technology is constantly updated,the scope of attacks is expanding,and the losses caused by attacks are increasing,which makes it gradually become a hot topic of the recent research.Based on the recent popular deep learning technology,this paper improves and innovates new botnet traffic detection algorithm and the imbalance of Botnet data samples,the details are as follows.(1)This paper first analyzes the characteristics of P2 P and other new botnet,and proposes a botnet detection algorithm based on improved CNN-LSTM fusion.The algorithm transforms the network flow into two-dimensional matrix,uses convolutional neural network to extract spatial features to capture its own characteristics,and then uses LSTM to extract temporal features from the data packet sequence to capture context,and finally uses feature fusion for recognition.At the same time,it improves the convolutional neural network structure,adopts small convolution kernel and highperformance Gelu activation function,so as to improve the detection accuracy of botnet.(2)In order to solve the problem of unbalanced data set caused by the difficulty of new botnet data collection and less samples,a botnet detection algorithm based on generative countermeasure network is proposed.By virtue of the strong generating ability of generative countermeasure network,the algorithm can be used as a data set expansion and enhancement in training,and the discriminator can effectively extract the features of the image in the training process.Finally,using the idea of transfer learning,the discriminator is modified and transferred to be a new botnet detector.Experiments show that the algorithm can accelerate the convergence speed and improve the detection accuracy under the condition of unbalanced data sets.The two improvements proposed in this paper can better complete the detection and identification of Botnet.Experiments show that the algorithm has good performance in both botnet detection and unbalanced data sets.Finally,this paper puts forward the shortcomings of the research and the direction of future improvement.