Machine Learning Classifiers for Network Intrusion Detection System: Comparative Study

Abstract

Network security risks are increasing at an exponential rate as Internet technology advances. Keeping the network protected is one of the most challenging of network security. Many security mechanisms were implemented to detect and identify any malicious activity on the network. Intrusion Detection System (IDS) one of the most used mechanisms to reduce the effects of these risks. Machine Learning (ML) classifiers are widely begin used to classify the network traffic as normal or abnormal. In this paper a comparative evaluation of the following ML classifiers: LogisticRegression, Multinomial Naive Bayesian, Gaussian Naive Bayesian, Bernoulli Naive Bayesian, k-Nearest Neighbors, Decision Tree, Adaptive Boosting, Random Forest, Multilayer Perceptron, and GradientBoosting is performed to specify the best classifier in identifying intrusion detection. The used evaluation metrics are accuracy, precision, and F-measure. The UNSW-NB15 dataset is used to assess ML classifiers. The experimental results show that the RandomForest classifier outperforms the other classifiers in terms of accuracy at 87%, precision 98%, and F-measure 84%.