Institutional Repository of Vilnius University
Darbuotojų keliamų vidinių grėsmių identifikavimas naudojant dirbtinį intelektą /
Abstract
dc:descriptionInsider threats pose a critical security challenge within modern organisations, as they originate from users who already possess legitimate access to organisational resources. This bachelor thesis investigates the application of artificial intelligence methods to insider threat detection, combining user behavioural analysis with textual data classification. Based on a systematic literature review, autoencoders, long short-term memory networks, and large language models were identified as the most promising approaches for this problem. Anomaly detection models were applied to user behavioural metrics. To evaluate whether model combinations yield better results than individual anomaly detection models, different models were combined using ensemble methods and a decision tree. Large language models of varying sizes were compared in terms of precision and computational resource usage to select the most suitable model for textual analysis. The best overall results were achieved by a hybrid system combining an ensemble of an autoencoder and a long short-term memory network with a large language model. A rule-based system for detecting SMTP tunnel-proxy based data exfiltration was additionally developed and tested. The results of this bachelor thesis indicate that hybrid artificial intelligence systems combining anomaly detection with large language models offer a viable approach to insider threat detection.
Degree
thesis:*- Grantor dc:publisher
- Institutional Repository of Vilnius University
- Year dc:date
- 2026
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Podbielski, Danielius,
Rights
dc:rights- Statement dc:rights
-
- info:eu-repo/semantics/openAccess
- Language dc:language
- lit
Identifiers
dc:identifier.*- Repository record dc:identifier
- https://repository.vu.lt/VU:ELABAETD308338477&prefLang=en_US