Intrusion Detection System Using Deep Learning Asymmetric Autoencoder (DLAA)

Abstract

To protect a network security, a good network IDS is essential. With the advancement of science and technology, present intrusion detection technology is unable to manage today's complex and volatile network abnormal traffic without taking into account the detection technology's scalability, sustainability, and training time. A new deep learning method is presented to address these issues, which used an unsupervised non-symmetric convolutional autoencoder to learn the dataset features. Furthermore, a novel method based on a non-symmetric convolutional autoencoder and a multiclass SVM is proposed. The KDD99 dataset is used to create the simulation. In comparison to other approaches, the experimental outcomes suggest that the proposed approach achieves good results, which considerably lowers training time and enhances the IDS detection capability.