A Supervised Hybrid DNN for Real-time Intrusion Detection in Firewalls

Abstract

With networking infrastructure and communication technology developments, Internetworking has become inalienable to everyone's day-to-day lives. The volume of utilization of computer networking has increased due to various factors that have enabled the modernization of cyber-physical systems, which have a remarkable effect on the networks' security. Securing the cyber-physical systems has become imperative, as internet utilization has reached record levels in these pandemic times. Various experimental studies have been performed in the Intrusion Detection System using ensemble machine learning algorithms on benchmark datasets. This article proposes a hybrid approach for a Network Intrusion Detection system using an ensemble Deep Neural Network (DNN). The proposed architecture classifies the normal and intruder packets from the real-time network packet traces. The architecture is optimized for analyzing the packets in an online network that can instantaneously produce the packet data classification result. The proposed DLNN is evaluated with real-time network traces and benchmark datasets to prove concept reliability, and scalability measures. The F1-score for the various ensemble Machine Learning Techniques such as Decision Tree Classifiers, Random Forest Classifier, and XG Boost are in the range of 87%. Our proposed method produces an F1-score of 94.8% for the real-time packet traces.