2026-04-17T17:49:05.974Z http://ezaposleni.singidunum.ac.rs/rest/sciNaucniRezultati/oai

ezaposleni.singidunum.ac.rs/rest/sciNaucniRezultati/oai:1:10115 2024-12-03T20:10:47Z 1

Insider Threat Detection Using Bidirectional Encoder Representations From Transformers and Optimized AdaBoost Classifier 2024 http://ezaposleni.singidunum.ac.rs/rest/sciNaucniRezultati/oai/record/1/10115 https://ieeexplore.ieee.org/document/10616526 K. Kumpf M. Protic L. Jovanovic M. Cajic M. Zivkovic N. Bacanin Cybersecurity plays an increasingly vital role in contemporary times, particularly as large companies face escalating threats and substantial losses, with ransomware attacks standing out as one of the more sophisticated forms of intrusion. In such attacks, hackers infiltrate systems, encrypt crucial company data, and then demand a ransom for data decryption or the restoration of control over the system. The application of natural language processing (NLP) emerges as a promising avenue, offering the potential to analyze attack patterns and recognize malicious intentions before actual attempts occur. Given the inherent specificity of each system, fine-tuning AI algorithms with special settings becomes imperative for optimal efficiency. This research delves into the synergy of NLP and robust classification algorithms to detect cyber attacks through website content analysis. Recognizing that classifier performance hinges on the judicious selection of hyperparameters, we explore a modified version of the recently introduced red fox algorithm (RFO). This adaptation is combined with bidirectional encoder representations from transformers to identify malicious intent in email content. The proposed approach undergoes testing on a real-world dataset, with the best models demonstrating an accuracy of 0.975822 suggesting viability. conferenceObject 1 6 10.1109/ICCSC62074.2024.10616526 2024 International Conference on Circuit, Systems and Communication (ICCSC), IEEE, Fes, Morocco