Anomaly detection In IoT using machine learning algorithms

Abstract

Internet of Things (IoT) is growing as one of the fastest developing technologies around the world. With IPv6 settling down, people have a lot of addressing spaces left that even allows sensors to communicate with each other while collecting data, leaving alone cars that communicate while travelling. IoT (Internet of Things) has changed how humans, machines and devices communicate with one another. However, with its growth, a very alarming topic is the security and privacy issues that are encountered regularly. As many devices exchange their data through internet, there is a high possibility that a device may be attacked with a malicious packet of data. In such cases, the security of the network of communication should be strong enough to identify malicious data. In other words, it is very important to create an intrusion detection system for the network. In our research, we propose a comparison between di erent machine learning algorithms that can be used to identify any malicious or anomalous data and provide the best algorithm for two data-sets. One dataset is on the environmental characteristics collected from sensors and another one is network dataset. The rst data-set is developed from the data exchanged between the sensors in an IoT environment and the second dataset is UNSW-NB15 data which is available online.