Network intrusion detection system using an optimized machine learning algorithm

Abstract

The rapid growth of the data-communications network for real-world commercial applications requires security and robustness. Network intrusion is one of the most prominent network attacks. Moreover, the variants of network intrusion have also been extensively reported in the literature. Network Intrusion Detection Systems (NIDS) have already been devised and proposed in the literature to handle this issue. In the recent literature, Kitsune, NIDS, and its dataset have received approx. 500 citations so far in 2019. But, still, the comprehensive parametric evaluation of this dataset using a machine learning algorithm was missing in the literature that could submit the best algorithm for network intrusion attack detection and classification in Kitsune. In this connection, two previous studies were reported to investigate the best machine algorithm (these two studies were reported by us). Through these studies, it was concluded that the Tree algorithm and its variants are best suited to detect and classify all eight types of network attacks available in the Kitsune dataset. In this study, the hyper-parameter optimization of the optimized Tree algorithm is presented for all eight types of network attack. In this study, the optimizer functions Bayesian, Grid Search, and Random Search were chosen. The performance has been ranked based on training and testing accuracy, training and testing cost, and prediction speed for each optimizer. This study will submit the best point hyper-parameter for the respective epoch against each optimizer.