Deep learning model for intrusion detection system utilizing convolution neural network

Abstract

An integral part of any reliable network security infrastructure is the intrusion detection system (IDS). Early attack detection can stop adversaries from further intruding on a network. Machine learning (ML) and deep learning (DL) techniques to automate intrusion threat detection at a scale never previously envisioned have snowballed during the past 10 years. Researchers, software engineers, and network professionals have been encouraged to reconsider the use of ML techniques, notably in cybersecurity. This article proposes a system for detecting intrusion with two approaches, the first utilizing a proposed hybrid convolutional neural network (CNN) and Dense layers. The second utilizes naïve Bayes (NB) ML techniques and compares the two approaches to determine the best detection accuracy. The preprocessing of network data is necessary. The suggested technique is evaluated using the UNSW-NB15 Dataset to create a reliable classifier and an effective IDS. The experimental results for the proposed CNN-dense classifier outperformed the ML and DL models. CNN has a 99.8% accuracy rate compared to previous studies. At the same time, the Gaussian naïve Bayes, which is considered the best among the ML-utilized classifiers, yielded an 83% accuracy rate.