Deep Learning-Based Network Intrusion Detection Using CNN and Enhanced UNSW-NB15 Multi-Class Dataset

Abstract

The need for sophisticated intrusion detection systems (IDS) has grown owing to the significant security risks and network attacks brought on by the proliferation of network devices or nodes such as computers, laptops, smart phones and others on the internet for data and information exchange. Deep learning has proven to be quite effective in a variety of disciplines and can handle large amounts of data. Security experts are therefore working to incorporate deep learning into an intrusion detection system. Multiclass model is designed for classifying network traffic into multiple attack categories. Several studies have been conducted on this subject, resulting in a wide variety of methods. Most of these approaches use predefined features extracted by an expert in order to classify network traffic. In addition, the UNSW-NB15 dataset was created in several distinct files and labelled using binary classification. The goal of this study is to separate the entire dataset into nine multiclass models and determine the highest network attacks from the dataset. The study examined how well deep learning performed in two classification categories (Binary and Multi-Class) using the improved UNSW-NB 15 dataset. The findings of the study discovered that Generic and Exploit Network Attacks from UNSW-NB 15 dataset contained highest network attacks. The model's training accuracy increases gradually while the validation accuracy improves up to 96.04% in multi-class classification.