3. Data-driven IoT Security Using Deep Learning Techniques

By Astaras Stefanos | Nikos Kefalakis | Angela-Maria Despotopoulou | John Soldatos

Downloaded: 992 times

Published: 30 Jun 2020

© 2020 Astaras Stefanos | Nikos Kefalakis | Angela-Maria Despotopoulou | John Soldatos

Abstract

For nearly two decades, machine learning techniques have been extensively researched in terms of their ability to indicate cybersecurity attacks in the networking and cloud assets of IT systems. To a lesser extent, they have been deployed in cybersecurity for IoT systems. Nevertheless, their potential for detecting attacks in all layers and components of an IoT system has not been adequately investigated. This chapter introduces deep learning techniques for the detection of abnormal behaviors and anomalies in the operation of state-of-the-art IoT systems that comprise smart objects such as connected vehicles and socially assisted robots. The presented approach uses auto-variable encoders and has been validated based on real-life datasets derived from smart objects and with very promising results. Moreover, the introduced algorithms have been integrated within a data-driven IoT security platform, namely the Secure IoT platform that has been presented in a previous chapter. Overall, the present chapter discusses a novel deep learning approach for IoT security, while validating it in realistic datasets other than the legacy openly available datasets that are commonly used by security researchers.