Deep learning and software-defined networks: Towards secure IoT architecture

Abstract

Internet of Things (IoT) introduces new challenges to conventional communication model. IoT networks characteristics, such as objects heterogeneity and scalability, require revolutionary solutions. Currently, there is no universal architecture for IoT. However, several architectures were proposed based on Software Defined Networks (SDN). SDN introduces network programmability, and centralisation, these features facilitate network abstractions, simplifying network management and eases evolution. In this paper, we investigate SDN as a novel communication architecture for IoT networking to enhance the security and resilience of IoT. SDN enhances network resilience and scalability which are essential in large-scale IoT deployments, e.g., smart cities. However, security is a significant concern for IoT while SDN deepens these concerns. SDN itself presents new security threats; specifically, threats related to the controller.We propose a secure, framework for IoT based on SDN. The framework is generalization for the integration of SDN and IoT. We focus on massive IoT deployment, for instance, smart cities applications, where, security is critical, and network traffic is enormous. The study investigates the SDN architecture from a security perspective. Improving SDN security will boost the deployment of SDN-based IoT architecture.We deploy an Intrusion detection system based on Deep Learning (DL). The detection module uses Restricted Boltzmann Machines (RBM). The precision rate shows significant improvements over standard ML, e.g. SVM and PCA.