Machine Learning Anomaly-Based Network Intrusion Detection: Experimental Evaluation
Résumé
The use of Machine Learning (ML) approaches to design anomaly-based network intrusion detection systems (A-NIDS) has been attracting growing interest due to, first, the ability of an A-NIDS to detect unpredictable and previously unseen network attacks, and second, the efficiency and accuracy of ML techniques to classify normal and malicious network traffic compared to other approaches. In this paper, we provide a comprehensive experimental evaluation of various ML approaches including Logistic Regression (LR), Decision Tree (DT), Random Forest (RF), and Artificial Neural Network (ANN), on a recently published benchmark dataset called UNSW-NB15 considering both binary and multi-class classification. Throughout the experiments, we show that ANN is more accurate and has fewer false alarm rates (FARs) compared to other classifiers, which makes Deep Learning (DL) approaches a good candidate compared to shallow learning for future research. Moreover, we conducted our experiments in a way to be served as a benchmark results since our used approaches are trained and tested on the configuration deliberately provided by the authors of UNSW-NB15 dataset for the purpose of direct comparison.
Origine | Fichiers produits par l'(les) auteur(s) |
---|