Research on the application of deep auto-encoder network in intrusion detection

The network data in the current network environment is enormous, complex, and multidimensional, which is hugely different from the past. The traditional machine learning method needs to manually extract a large number of features in the face of complex high-dimensional data, and the process is complex and computationally intensive, which is not conducive to the current real-time and accuracy requirements of intrusion detection. Thus, in order to reduce the data dimension and eliminate redundant information, an intrusion detection method based on DAN-BP which combines deep auto-encoder network (DAN) and BP algorithm is proposed. First, a DAN model was constructed by overlaying several auto-encoder networks, and the network feature data was used as the input of the model, which enables the model to intelligently extract the distribution rules of the network data layer by layer, thereby obtaining a new low-dimensional feature data set. Then the low-dimensional data was classified and identified by the BP algorithm. In this research, the regularization correction was added to the auto-encoder network to prevent the trained auto-encoder network from directly copying the input information and influencing the training effect. Moreover, noise was added to the input data, and the reconstruction error of the original data and the output data was learned to achieve the purpose of denoising so that the learned new feature data is more robust. The traditional dimensionality reduction method and the proposed intrusion detection method were compared in this paper. Results show that the proposed method has better performance in classification accuracy, false alarm rate, and detection rate. © 2019, Editorial Board of Journal of Harbin Institute of Technology. All right reserved.