Democritus University of Thrace (DUTH)
Machine learning with applications in cyber security
Abstract
dc:descriptionIn the last few years, Machine Learning has had a revolutionary impact in the cyber security domain. This integration was particularly enhanced by the emergence of highly sophisticated cyber threats, whose detection and mitigation are getting harder to achieve using traditional approaches alone, and Machine Learning’s ability to provide tools for intelligent threat detection, such as big data analysis, anomalous behaviour detection, and real-time prediction. At the same time, recent years have witnessed an increasing number of cyber-attacks that are specifically designed for disrupting critical infrastructure; such as telecommunications, healthcare, energy, and public services, whose continuous provision of high-availability services is of paramount importance. Additionally, one primary problem for many Intrusion Detection Systems when dealing with emerging threats is their heavy reliance on known signatures, an approach that has been proven to exhibit poor results for detecting zero-day attacks. By adopting an application-centric approach, this thesis addresses these deficiencies and proposes a series of defensive tools that integrate the effectiveness of Machine Learning inference as their core element. In doing so, it also deals with the inherent computational challenges that characterise Machine Learning-based processes and includes as many challenges as possible for each security problem to explore each detection application in its most realistic scenario. Emphasis is also given on the protection of critical infrastructure, an environment that is particularly vulnerable to attack operations. Additionally, it incorporates fuzzy logic principles into the proposed solutions, acknowledging their vast contribution to the field of machine intelligence. The contributions of this work include a) the use of DNS data to mitigate overload states of the DNS servers, b) the investigation of an underestimated attack that targets temperature-sensitive infrastructures such as data centres, and c) an extended analysis of malicious activity detection (which was conducted in two parts) that is based on image conversion, intelligent colour assignment, space-filling curves, Deep Learning, and economical image representation. While the first two applications of this work contribute immediately to intrusions targeting critical systems, the application based on binary visualisation and its extension can be applied to critical infrastructure too. Finally, all applications can be seen as vital parts of a modular Intrusion Detection System that promotes intelligent malicious activity detection based on historical data.
Degree
thesis:*- Grantor dc:publisher
- Democritus University of Thrace (DUTH)
- Year dc:date
- 2024
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Saridou, Despoina
- Σαρίδου, Δέσποινα
Subjects
dc:subject × 31- Μηχανική μάθηση
- Κυβερνοασφάλεια
- Νευρωνικά δίκτυα
- Ανίχνευση κακόβουλου λογισμικού
- Ασαφής λογική
- Κρίσιμες υποδομές
- Συστήματα ανίχνευσης εισβολών
- Βαθιά μάθηση
- Machine learning
- Cyber security
- Cybersecurity
- Neural networks
- Malicious software detection
- Fuzzy logic
- Critical infrastructures
- Intrusion detection systems
- Deep learning
- Επιστήμη Ηλεκτρονικών Υπολογιστών και Πληροφορική
- Φυσικές Επιστήμες
- Τεχνητή νοημοσύνη
- Επιστήμη ηλεκτρονικών υπολογιστών και Πληροφορική, άλλοι τομείς
- Μαθηματικά
- Μαθηματική λογική
- Στατιστική και Πιθανότητες
- Computer and Information Sciences
- Natural Sciences
- Artificial Intelligence
- Computer and Information sciences, miscellaneous
- Mathematics
- Logic
- Statistics and Probability
Rights
- Language dc:language
- eng
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/57541