Ανίχνευση ανωμαλιών Zero-day με χρήση AutoEncoders

Μπαζώτης, Νικόλαος; Bazotis, Nikolaos

dc.contributor.author	Μπαζώτης, Νικόλαος	el
dc.contributor.author	Bazotis, Nikolaos	en
dc.date.accessioned	2024-01-11T08:32:22Z
dc.date.available	2024-01-11T08:32:22Z
dc.identifier.uri	https://dspace.lib.ntua.gr/xmlui/handle/123456789/58563
dc.identifier.uri	http://dx.doi.org/10.26240/heal.ntua.26259
dc.rights	Default License
dc.subject	Νευρωνικά Δίκτυα	el
dc.subject	Μηχανική Μάθηση	el
dc.subject	Σύστημα Ανίχνευσης Διείσδυσης	el
dc.subject	Επιθέσεις Zero-day	el
dc.subject	Μη επιβλεπώμενη μάθηση	el
dc.subject	Deep learning	en
dc.subject	Neural nets	en
dc.subject	Cyber security	en
dc.subject	Intrusion Detection Systems	en
dc.subject	Unsupervised learning	en
dc.title	Ανίχνευση ανωμαλιών Zero-day με χρήση AutoEncoders	el
heal.type	bachelorThesis
heal.secondaryTitle	Zero-day Anomaly Detection using AutoEncoders	en
heal.generalDescription	Intrusion Detection System using AutoEncoders for zero-day attack anomaly detection	en
heal.classification	Deep Learning	en
heal.classification	IT systems	en
heal.classification	Communications	en
heal.classification	Intrusion Detection System	en
heal.language	en
heal.access	campus
heal.recordProvider	ntua	el
heal.publicationDate	2023-07-10
heal.abstract	Zero-day attacks have become increasingly sophisticated in recent years, presenting a significant threat to both businesses and individuals. The danger lies in these attacks exploiting unknown vulnerabilities in software or hardware, making traditional security measures often ineffective. Consequently, organizations are left vulnerable to data breaches, financial losses, and reputational damage. In the wake of this escalating threat, cyber security researchers and practitioners are turning to deep learning techniques such as AutoEncoders. Particularly, the focus is on unsupervised learning, which relies on unlabeled data. This approach circumvents the challenges and potential inaccuracies associated with data labeling, thereby offering a more effective defense mechanism against zero-day attacks. This diploma thesis sets out to implement a Network Threat Detection System using AutoEncoders, targeting early detection of zero-day attacks before they can be exploited. The proposed system will leverage the CICIDS 2017 dataset, a well-recognized benchmark for evaluating intrusion detection systems, encompassing network traffic data that simulates diverse types of attacks, including zero-day attacks. Training the deep learning model on this dataset enables the system to identify anomalies in network traffic data and flag potential zero-day attacks. An integral part of this work involves experimenting with different kinds of AutoEncoders, such as Denoising AutoEncoders, Variational AutoEncoders (VAEs), Deep AutoEncoders, and Beta VAEs. A comparative study is carried out between these models, and their performance is measured against shallow models from existing literature, specifically SVM in this case. Through such rigorous experimentation and comparison, this thesis aims to advance the current understanding and application of deep learning techniques in preventing zero-day attacks.	en
heal.advisorName	Maglaris, Vasilis	en
heal.committeeMemberName	Stamou, Giorgos	en
heal.committeeMemberName	Παπαβασιλείου, Συμεών	el
heal.committeeMemberName	Σύκας, Ευστάθιος	el
heal.academicPublisher	Εθνικό Μετσόβιο Πολυτεχνείο. Σχολή Ηλεκτρολόγων Μηχανικών και Μηχανικών Υπολογιστών. Τομέας Επικοινωνιών, Ηλεκτρονικής και Συστημάτων Πληροφορικής	el
heal.academicPublisherID	ntua
heal.numberOfPages	126 σ.	el
heal.fullTextAvailability	false