Intrusion Detection for IoT Environments Through Side-Channel and Machine Learning Techniques

The rise of the Internet of Things (IoT) technology during the past decade has resulted in multiple applications across a large variety of fields. Some of the data processed using this technology can be specially sensitive, and the devices involved can be prone to cyberattacks, which has resulted in a rising interest in the field of information security applied to IoT. This study presents a method for analyzing an IoT network to detect attacks using side-channel techniques that monitor the power usage of the devices. It shows that it is possible to employ a monitoring system powered by Machine Learning to detect intrusions without interfering with the normal behavior of the devices. Tests yield positive results under a range of scenarios, including using a custom dataset, detecting new attacks previously unseen by the models, and detecting attacks in real time. The main advantages of the proposed system are simplicity, reproducibility (both code and data are made available) and portability, since it can be deployed on all kinds of devices and does not have a high demand of resources. Several deployment strategies are proposed, depending on the structure of the target IoT network and the power constraints of the devices.