Enhancing anomaly detection through restricted Boltzmann machine features projection
- PDF / 2,006,362 Bytes
- 9 Pages / 595.276 x 790.866 pts Page_size
- 5 Downloads / 293 Views
ORIGINAL RESEARCH
Enhancing anomaly detection through restricted Boltzmann machine features projection Gustavo H. de Rosa1
•
Mateus Roder1 • Daniel F. S. Santos1 • Kelton A. P. Costa1
Received: 11 April 2020 / Accepted: 30 September 2020 Bharati Vidyapeeth’s Institute of Computer Applications and Management 2020
Abstract Technology has been nurturing a wide range of applications in the past decades, assisting humans in automating some of their daily tasks. Nevertheless, more advanced technology systems also expose some potential flaws, which encourage malicious users to explore and break their security. Researchers attempted to overcome such problems by fostering intrusion detection systems, which are security layers that try to detect mischievous attempts. Apart from that, increasing demand for machine learning also enabled the possibility of combining such approaches in order to provide more robust detection systems. In this context, we introduce a novel approach to deal with anomaly detection, where instead of using the problem’s raw features, we project them through a restricted Boltzmann machine. The intended approach was assessed under a well-known literature anomaly detection dataset and achieved suitable results, better than some state-of-theart approaches. Keywords Intrusion detection system Anomaly detection Machine learning Restricted Boltzmann machine
& Gustavo H. de Rosa [email protected] Mateus Roder [email protected] Daniel F. S. Santos [email protected] Kelton A. P. Costa [email protected] 1
Department of Computing, Sa˜o Paulo State University, Av. Eng. Luiz Edmundo Carrijo Coube, 14-01, Bauru, SP 17033-360, Brazil
1 Introduction The development of technology fostered the capacity of autonomous systems solving particular tasks. Nowadays, it is common to observe digital agents being applied everywhere, ranging from daily bank tasks to world wide web browsers. However, increasing connectivity among users also fosters mischievous intents, i.e., users who attempt to break digital security rules and gain access to unauthorized devices or confidential information [5]. An attempt to overcome such a problem is to employ digital barriers, commonly known as anomaly-based intrusion detection systems (IDS) [1]. Essentially, they are security layers that enable any application to detect a malicious attempt and revoke their access instantly [4]. Moreover, they can analyze previous data and decide whether an incoming attack is possible or not [19], helping systems to protect themselves against security attacks. Even though IDSs seems to be the ideal tool to cope with such attempts, some flaws need to be addressed [22]. Every day new attacks and security breaches are being explored, inflicting the necessity of updating the IDS knowledge. Furthermore, depending on the type of application the IDS is protecting, there is no possibility of raising false positives or false negatives. In the past decade, researchers are combining concepts of machine learning techniques and IDS in an a
Data Loading...
