2026-08-15T17:01:24.506Z http://ezaposleni.singidunum.ac.rs/rest/sciNaucniRezultati/oai

ezaposleni.singidunum.ac.rs/rest/sciNaucniRezultati/oai:3:12144 2026-08-10T21:23:29Z 3

The XGBoost Classifier Tuned with an Adapted Particle Swarm Optimization Algorithm for Detecting Anomalies in Operational Logs of Cloud-Based Systems, chapter in LNNS Lecture Notes in Networks and Systems: IDSCS 2025: Data Science and Security, Springer, volume 1947 2026 http://ezaposleni.singidunum.ac.rs/rest/sciNaucniRezultati/oai/record/3/12144 https://link.springer.com/chapter/10.1007/978-3-032-24360-7_12 M. Zivkovic N. Bacanin J. Maricic M. Antonijevic B. Radomirovic A. Petrovic S. Anetic V. Zeljkovic The reliance on cloud-based technologies is expected to intensify in the future, making scalability inseparable from the robustness of their security. This study focuses on analyzing system log files and introduces a framework aimed at identifying abnormal patterns within them. To achieve this, XGBoost classifier is employed for anomaly prediction, with preprocessing carried out using Word2Vec, a widely used natural language processing technique. The central innovation lies in a newly designed version of the particle swarm optimization (PSO) algorithm, tailored for fine-tuning XGBoost set of hyperparameters for the anomaly detection task. The effectiveness of this approach is benchmarked against several leading metaheuristic optimization methods. Experimental evaluation demonstrates that the proposed model achieves remarkable predictive accuracy, with top-performing configurations reaching nearly 98.48% accuracy in the test scenario. bookPart Springer, Cham 119 133 10.1007/978-3-032-24360-7_12 LNNS Lecture Notes in Networks and Systems: IDSCS 2025: Proceedings of International Conference on Data Science and Security, volume 1947