2026-09-26T15:24:27.798Z http://ezaposleni.singidunum.ac.rs/rest/sciNaucniRezultati/oai

ezaposleni.singidunum.ac.rs/rest/sciNaucniRezultati/oai:3:12174 2026-09-10T23:43:21Z 3

XGBoost Classifier Optimized with a Modified Sinh Cosh Algorithm for Anomaly Detection in Cloud System Operational Logs, Chapter in LNNS Lecture Notes in Networks and Systems: ICDPN 2025: Data Processing and Networking, Springer, volume 2005 2026 http://ezaposleni.singidunum.ac.rs/rest/sciNaucniRezultati/oai/record/3/12174 https://link.springer.com/chapter/10.1007/978-3-032-27691-9_21 S. Anetic I. Kosta B. Radomirovic N. Lukovic T. Zivkovic M. Zivkovic M. Antonijevic N. Bacanin As dependence on cloud infrastructures continues to expand, ensuring scalable yet secure systems becomes increasingly critical. In this context, the present study examines system log data and proposes a dedicated framework for detecting anomalous patterns. The methodology integrates the XGBoost algorithm for anomaly classification, with log data preprocessed through custom developed Gensim embeddings to transform raw text into meaningful numerical features. The primary contribution is the development of a customized variant of the well-established sinh cosh algorithm, specifically adapted to optimize XGBoost hyperparameters for detecting anomalies in log data. The performance of this tailored approach is systematically compared against several prominent metaheuristics optimization techniques. Experimental findings confirm its strong effectiveness, with the best configurations attaining a predictive accuracy of 100% on the test dataset. bookPart Springer, Cham 245 259 10.1007/978-3-032-27691-9_21 LNNS Lecture Notes in Networks and Systems: ICDPN 2025: Proceedings of International Conference on Data Processing and Networking, volume 2005