Aleksander Vanberg Eriksen

• Pal, Kim Kristoffer; Eriksen, Aleksander Vanberg & Dinh, Thi Thuy Nga (2025). XGBoost Feature Selection for Multi-Class and Binary Classification on UNSW-NB15 Dataset. In IEEE (Eds.), 2025 IEEE International Conference on Consumer Electronics (ICCE). IEEE (Institute of Electrical and Electronics Engineers). ISSN 9798331521165. doi: 10.1109/icce63647.2025.10930023. Full text in Research Archive Show summary

  Network intrusion detection systems (NIDS) are a critical part of ensuring the integrity and reliability of a network, which is essential in secure communication. To achieve this, an efficient NIDS with as low complexity as possible is preferred. Therefore, this paper presents a feature selection (FS) method that utilizes extreme gradient boosting (XGBoost) to train binary and multi-class machine learning (ML) algorithms on the 19 most important features in the UNSW-NB15 dataset. The FS method is evaluated on seven ML algorithms namely Artificial Neural Network (ANN), Logistic Regression (LR), k-Nearest-Neighbors (kNN), Support Vector Machine (SVM), Decision Tree (DT), Random Forest (RF), and XGBoost. Results from both the multi-class and binary classifications showed that RF and XGBoost had the highest accuracy and Fl-score at (77.14%, 78%), (87.445%, 87%), (78.156%, 79%) and (87.527%, 87%), respectively. Compared to existing works, most ML algorithms demonstrate better or similar performance on the UNSW-NB15 dataset. In addition, the proposed FS method significantly reduces the running time for XGBoost and RF by approximately 80% and 48%, respectively in both binary and multi-class classifications.

View all works in NVA