Machine Learning for Automated Industrial IoT Attack Detection: An Efficiency-Complexity Trade-off

Abstract

Critical city infrastructures that depend on smart Industrial Internet of Things (IoT) devices have been increasingly becoming a target of cyberterrorist or hacker attacks. Although this has led to multiple studies in the recent past, there exists a paucity of literature concerning real-time Industrial IoT attack detection. The goal of this article is to build a machine-learning approach using Industrial IoT sensor readings for accurately tracking down Industrial IoT attacks in real time. We analyze IoT system behavior under a lab-controlled series of attacks on a Secure Water Treatment (SWaT) system. The system is analytically challenging in that it results in sensor readings that resemble waveforms. To that end, we develop a novel early detection method using functional shape analysis (FSA) to extract features from the data that can capture the profile of the waveform. Our results show an efficiency-complexity trade-off between functional and non-functional methods in predicting IoT attacks.