Intrusion Detection System Using Deep Learning Asymmetric Autoencoder (DLAA)

Abstract

A good network intrusion detection system (IDS) is essential for protecting network security. Current intrusion detection technology is unable to manage abnormal traffic in today's complex and volatile network unless it takes scalability, sustainability, and training time into account. This chapter presents a new deep learning method for addressing these issues, which used an unsupervised non-symmetric convolutional autoencoder to learn the dataset features. A novel method based on a non-symmetric convolutional autoencoder and a multi-class support vector machine (SVM) is also proposed. The KDD99 dataset was used to create the simulation. Experimental outcomes suggest that the proposed approach achieves good results compared to other approaches, considerably reducing training time and enhancing the IDS detection capability.