Convolution Neural Network Machine Learning Algorithm Prediction Model for Intrusion Detection

Abstract

Software Defined Networking (SDN) is evolving as a brand-new approach to the growth and innovation of the Internet. Since SDN can offer controllable, dynamic, and affordable networking, it is anticipated to be the Internet's ideal future. A rare chance to achieve network security in a more effective and flexible way is presented by the introduction of SDN. Because it has centralised control, SDN has the advantage of better network security provisioning as compared to traditional networks. However, in order to increase SDN security, it is necessary to address a number of additional network security challenges brought about by the SDN architecture's flexibility. The centralised controller, the control-data interface, and the control-application interfaces are the SDN's original structural weaknesses. Intruders may take advantage of these weaknesses.to conduct several types of attacks. A crucial component of network architecture known as the Network Intrusion Detection System (NIDS) is utilised to identify network intrusions and secure the entire network. Using Deep Learning (DL) methods, we suggest an SDN-based NIDS (DeepIDS) in this thesis to look for anomalies in the SDN architecture. First, using various flow features, we assess the capability of DL for flow-based anomaly identification. We demonstrate through studies that the DL technique has the capacity to detect flow-based anomalies in the SDN context. We also suggest a Gated Recurrent Unit Recurrent Neural Network to boost DeepIDS's detection rate. Our experimental findings demonstrate that the suggested model considerably increases the detection rate without degrading network performance. The effectiveness of our system in terms of precision, throughput, latency, and resource utilisation demonstrates that DeepIDS does not negatively impact the OpenFlow controller's performance, making it a workable strategy. Finally, we present an unsupervised method to address the issue of an unlabelled and unbalanced dataset. This method results in a significant reduction in processing time while producing a high detection rate. Through thorough experimental evaluations, we determine that our suggested strategy we conclude that our proposed approach exhibits a strong potential for intrusion detection in the SDN environments.