Evaluating neural networks using Bi-Directional LSTM for network IDS (intrusion detection systems) in cyber security

Abstract

An Intrusion detection system is a fundamental layer incorporated in the network system. Due to enormous amount of traffic in the Network, the attacker waits for the chance to cause massive damage to the network and the network users. Even by using IDs the network admin face difficulties in identifying threats, attacks, and vulnerabilities in existing methods. This paper focuses on applying the deep learning method and an IDs model based on Bi-directional LSTM. KDDCUP-99 and UNSW-NB15 datasets are used in experiments to test the designed system. The model using Bi-directional LSTM gave outstanding results with 99% accuracy for both KDDCUP-99 and UNSW-NB15 datasets. The work was repeated by varying the activation functions used in the network. For both the datasets, softmax and relu gave impressive results with an average of 99.5% accuracy. The results were compared with the state-of-the-art methods. From the comparison we can conclude that Bi-directional LSTM performs better compared other related works in the literature.