Evaluation of Ensemble and Deep Learning Classifiers on CSE-CIC-IDS2018 Dataset for Intelligent NIDS

Kaushik Datta; Tapas Samanta; Sarbajit Pal

Abstract

Network Intrusion Detection System (NIDS) plays an active role in preventing cyberattacks by early detection of threats before it really starts affecting targeted information services. Over the years, many intrusion detection system (IDS) have been developed applying signature or rule-based approach to prevent unauthorised access of network or computer devices. However, ever growing landscape of cyberattacks in recent years has motivated present day researchers to design and develop more accurate IDS using modern Machine Learning (ML) methods which identify attacks through anomaly detection. Development of intelligent NIDS highly depends on a rich, up-to-date and contemporary dataset which consists of relevant attributes and real-world scenario of cyberattacks. Varity of datasets are available for this purpose among which KDDCUP99, NSLKDD, ISCX2012, CICIDS2017, CICIDS2018, Kyoto etc. are the most popular ones and widely used. This study reports our observations on the performance of two well-known classifiers among Ensemble Learning methods, namely Random Forest and XGBoost and of Deep Neural Network classifier on the CSE-CIC-IDS2018 dataset which is relatively a new one and covers many contemporary cyberattacks. Their performances are evaluated using multiple metrics including Precision-Recall curve which has been proved to be more useful in case of imbalanced dataset like CSE-CIC-IDS2018.

Abstract

Keywords