Network Intrusion Detection CNN Model for Realistic Network Attacks Based on Network Traffic Classification
Alla Abd El-Rady, Heba Osama, Rowayda Sadik, Hesham El Badwy
Abstract
Network intrusion detection has an important role in providing security to networks and computer systems. It applies different artificial intelligence technologies in order to improve performance against various cyber-attacks. Applying Deep Learning (DL) techniques has a considerable impact compared to using traditional Machine Learning (ML) methods. Recently, Convolutional Neural Network (CNN) has been widely used by researchers to enhance Intrusion Detection Systems (IDSs). This paper aims to build a customized CNN model to improve the accuracy of IDSs. The study involves comparing the results obtained from the proposed CNN model with those obtained from Random Forest which is a well-known machine learning technique utilized frequently in IDSs. The performance of both models was evaluated using standard measurements such as accuracy and F-measure. Two datasets were used, UNSW-NB15 and CSE-CICIDS2018, to demonstrate the efficacy of our proposed model. The proposed CNN model achieved better results than the Random Forest algorithm. A comparison with existing recent IDSs was carried out. The result of this study provides insights and proves the effectiveness of using CNNs for IDSs and helps in identifying the best approach for building efficient and accurate CNN based IDSs with the accuracy of 99.18% using CSE-CICIDS2018 dataset and 99.70% using UNSW-NB15 dataset.