Intelligent SDN to enhance security in IoT networks

Abstract

Software-defined networking (SDN) is a revolutionary technology that has revolutionised network management by providing flexibility and adaptability. As the popularity of SDN increases, it is crucial to address security vulnerabilities in these dynamic networks. This paper proposes a framework for enhancing security in SDN by utilising three separate Deep Learning models, namely Deep Neural Network (DNN), Convolutional Neural Network (CNN), and Long Short-Term Memory (LSTM). This framework is utilised for the InSDN dataset, a huge dataset specifically created for SDN security research. The dataset consists of a total of 343,939 instances, encompassing both normal and attack traffic. The regular data yields a sum of 68,424, whereas the attack traffic comprises 275,515 occurrences. This study employs multiclassification algorithms to precisely detect and categorise diverse security threats in SDN. The InSDN dataset faces issues related to class imbalance, which are addressed by using the Synthetic Minority Over-sampling Technique (SMOTE). The SMOTE technique is utilised to create artificial instances of the underrepresented class, hence achieving a more equitable distribution of security hazards within the dataset. This strategy improves the efficacy of multiclassification techniques, ultimately resulting in greater accuracy in the identification and classification of different security threats in SDN environments. The initial DNN model exhibited satisfactory performance, with an accuracy of 87%. The second CNN model demonstrated strong and consistent performance, with an accuracy rate of 99%. In addition, an LSTM model attained a 90% accuracy rate.