An Effective Recurrent Neural Network (RNN) based Intrusion Detection via Bi-directional Long Short-Term Memory

Abstract

The evolution of communication and information systems has raised the volume of data distributed through the internet. As an effect, a majority of digital resources have been increased, so does the challenge of cybersecurity. Intrusion detection systems (IDSs) are closely connected to a holistic approach for preventing cyberattacks. Due to the high utilization of network traffic in the cyber world, conventional machine learning approaches used in intrusion detection systems are becomes ineffective. Recently evolved deep learning techniques are successfully applied in the detection and classification of threats at both the network and host levels, with a focus on deep learning. This study proposed an efficient IDS based on Recurrent Neural Network (RNN) via Bi-directional Long Short- Term Memory (RNN BiLSTM). The strategy uses a two-step mechanism to develop the expertise of the suggested solution to address network problems. This research aims to determine the algorithm’s processing time and increase attack classification accuracy. The proposed model was evaluated on the CICIDS2017 intrusion detection dataset. The Random forest and Principal component analysis algorithms were used to detecting the valuable features and eliminating the unwanted features from the given dataset. The findings revealed that BLSTM outperform all other RNN architectures in terms of classification accura 98.48%.