An Intelligent Intrusion Detection System Using Hybrid Deep Learning Approaches in Cloud Environment

Abstract

An Intrusion Detection System (IDS) detects suspicious activities and sends alerts when they are found. Based on these alerts, the issue is investigated, and appropriate actions are taken to remediate the threat. The traffic in a network is examined by a network-based intrusion detection system using various traffic tools that collect and analyse traffic data utilizing detection algorithms. Virtualization is used to construct the cloud infrastructure, which renders the virtual network flow between the virtual machines and it is mostly unidentifiable by typical intrusion detection systems. Previous studies proposed a software-defined network technology to reroute network traffic to a Snort IDS for detection of malicious attacks. However, this is incapable of detecting unknown attacks and adapting to large-scale traffic. Deep learning algorithms are used automatically to extract essential features from raw network data, which can then be fed into a shallow classifier for effective malicious attack detection. The main objective of the proposed system is to utilize a combination of a sparse autoencoder and stacked contractive autoencoder (S-SCAE) along with a Bi-DLDA (Bi-directional LSTM followed by a dense layer, a dropout layer, and a layer with attention mechanism) for detecting intrusions in a cloud environment. Moreover, a cloud intrusion detection system that designed to collect the data traffic from the NSL-KDD dataset and applies the S-SCAE + Bi-DLDA algorithm to determine if the received packet is malicious or non-malicious. To assess the proposed system's detection performance, a variety of measures were used such as precision, recall rate, and accuracy. The proposed model achieves precision, recall rate, and accuracy of 99%, 98%, and over 98% respectively, according to simulation findings.