Enhanced Intrusion Detection and Mitigation in SDN Based IoT Networks Using Deep Learning and Explainable AI Techniques

Abstract

The increasing number of IoT applications in critical industries, such as newlinehealthcare, has generated significant security concerns, necessitating the newlineimplementation of robust Intrusion Detection Systems (IDS). Conventional IDS newlinemethodologies are inefficient in addressing particular challenges, including traffic flow newlinedynamics, resource constraints, and scalability difficulties. The current research proves newlinethat the SDN-IoT based IDS is more effective than traditional systems providing newlinecentralized management, programmability, and capacity to manage dynamic network newlinetraffic. The centralised architecture of SDN facilitates intelligent operations, resulting newlinein immediate response to attacks, hence enhancing efficiency and security. newlineThe proposed SDN-IoT based Intrusion Detection System provides newlineautomated feature learning and extraction using enhanced Deep Learning model. These newlineextracted features are sent to the Erk-herd hyperparameter optimization for optimum newlinefeature selection. These optimal features are sent to the XGBoost model for attack newlinedetection. The proposed hybrid model depicts an F1 score of 0.93. The automated newlinefeature selection reduces the training time of the proposed IDS framework and since newlinemachine learning is used for attack detection, the overhead problem is reduced. Also, newlinesince the collected dataset is imbalanced, the application of machine learning newlineclassification reduces the overfitting issues newline