Hybrid Variational Autoencoders-DenseNet Convolution Neural-Network Based Deep Learning Approach for Intrusion Detection System

Abstract

In the modern big data environment, a deep learning approach for intrusion detection was seen as more effective and efficient to counter known as well as unknown attacks. Much research has been conducted on intrusion detection systems using a deep learning approach, as well as a hybrid model of deep learning, and continues to be conducted to minimize the risk of intrusion. In this paper hybrid approach is used using variational autoencoders and DenseNet convolutional neural networks to model an intrusion detection system. An experiment is performed using the NSL-KDD dataset to calculate the performance metrics of the model, which contains four types of attacks: probe attack, DoS attack, user-to-root attack, and remote to local attack. The result obtained is compared with the results of other related works. The result obtained from the experiment is quite satisfactory as well, and in comparison with other related works, the model seems comparatively efficient. An intrusion detection system using a VAE-DCNN hybrid model can detect malicious activities very efficiently and correctly, as the performance metric of this model is quite good. The possibility of a false alarm will be reduced using this hybrid model, as the false alarm rate is low.