Article Preview
Top1. Introduction
The network prevention system is increasing with the advancement of science and technology, but it still confronts many obstacles to invade attack. According to Symantec's 2018 report on Internet Security Threats, the incidence of hostile network incursions increased by 200 percent in 2018, and the growth rate of assaults against the Internet of Things (IoT) is even greater at 600 percent, which is quite disturbing and dangerous to people's interests. In cyberspace, space security plays a vital role in protecting the network. However, in today's sophisticated networks warfare, cyberspace security faces many new obstacles. Such as Firstly, the popularity of modern IoT and constant usage of cloud services have resulted in a significant growth in network data volume, and it’s expected that this trend will continue. Secondly, there are several new protocols added to the modern network traffic (Ji et. al. in 2020), which adds lots of difficulties and complexities in network security detection system. Therefore, it is necessary to adjust the network protection measures to adopt to the constantly changing network conditions.
The network firewall, which can monitor intrusion activities statically, is one of the most popular measures to defend network security. It employed the intrusion detection systems as a second line of defense for dynamic network security protection, and they can actively protect computer logs and system files changes to get the sign of attack. By reviewing log data, it can detect any changes in the files as evidence of an attack, such as unexpected traffic or an undiscovered, unknown attack. There are two types of intrusion detection system (IDS): host-based intrusion detection system (HIDS) and network intrusion detection system (NIDS). According to Chawdhary et. al. in 2017, the information gained by a single computer system is used by host-based IDS, whereas network-based IDS collect raw network traffic and analyze intrusion flags. It is now the biggest source of worry in improving the performance and efficiency as discussed by Kim et. Al in 2017.
ML and deep learning approaches are used in a variety of applications in recent years (Shen et. al. in 2017). A lot of research has been carried out to identify cyber intrusions and measuring methods have been researched with positive alarm accuracy. However, most existing intrusion detection systems still cannot detect the new aberrant traffic because of the difficulty in finding valid training data, the duration of training time, high error rates during evaluation. This strategy is extremely reliant on data availability, and it necessitates the use of human specialists to filter data and rely on their expertise, which is both time consuming and costly (Shano et. Al in 2018). As a result, the detection mechanism should be tweaked so that they can self-learn and detect intrusions with more accuracy.
In recent years, machine learning based k-nearest neighbor (KNN) and support vector machine (SVM) algorithms (Reaz et. al. in 2016, Dave et. al. in 2013) are applied in intrusion detection and showed virtuoso performance. But these algorithms have certain limitations i.e they can work on limited datasets and cannot solve complex problems. To overcome the problem of shallow learning, some researchers have shown that deep learning algorithms are better as compare to other intrusion methods (Hou et. al in 2016).
To answer the afore identified issues, this study aims to implement a new supervised non-symmetric convolutional autoencoder and support vector machine (SVM) intrusion detection method, combined with deep Learning and shallow classifiers, to evaluate and discuss data, adapt to changes in modern network traffic, and use KDD99 dataset. The testing results suggest that this strategy successfully increases intrusion detection, efficiency, and detection capabilities of IDS. The following are the primary contributions of this paper-
1.This study discuss the deep learning asymmetric autoencoder algorithm (DLAAA), it provides non-symmetric data reduction dimension, solved the deficiencies of CNN and auto-encoder (Ji et. al. in 2020). Therefore, with the deep belief network and stacked autoencoder (SAE) and other leading methods, the technology proposed in this article changes the classification results.