Intelligent Intrusion Detection System Using Hybrid Deep Learning Model of Parametric Swish-based Siamese Recurrent Highway Networks for Software Defined Networks

Software-defined Networking (SDN) is a future internet technology that provides a global network management for dynamically controlling the network flows.Compared with traditional networks, the centralized control of SDN provides better provisioning of network security, but the network intrusions of new patterns are manifested using the flexible SDN architecture.Therefore, developing intrusion detection systems (IDS) to strengthen SDN network security by improving anomaly detection is essential.While traditional IDS methods were ineffective for new unknown threats such as zero-day attacks, the Machine Learning (ML) and Deep Learning (DL) methods provided low false positives and improved detection rates.Yet, the slower training process in ML and high computational complexity in DL are significant problems for achieving better intrusion detection in SDN.This paper presents a lightweight DL model named Parametric Swish-based Siamese Recurrent Highway Networks (PS-SRHN) to detect different attacks and unknown patterns and increase the detection rates in SDN.Initially, the data preprocessing steps are applied, and the features are extracted from the datasets, the important features of which are selected based on the Pearson Correlation Coefficient (PCC).Then, the selected features are fed to the proposed PS-SRHN classifier model for final detection.This PS-SRHN is formulated by integrating Recurrent Highway Networks (RHN) in Siamese Network architecture and introducing the Parameter Swish activation function to learn the sigmoid coefficient's optimal value and reduce vanishing gradients.Validations showed that the proposed PS-SRHN classifier achieved 99.57%, 100%, and 100% accuracy and 17.45seconds, 11.33seconds and 21.12seconds processing time for NSL-KDD, SDN-IoT and InSDN datasets, respectively.