Detecting Application-Level Associations Between IoT Devices Using A Modified Apriori Algorithm

Abstract

Internet of Things (IoT) enabled devices are becoming increasingly common, but their reliance on internet connectivity reduces their overall reliability. The use of cloud servers is one way of achieving interoperability between different IoT devices. It is not necessary for two devices to know how to directly communicate with each other, because their vendor’s cloud servers will do it for them. As a result, cloud servers have become a critical part of the IoT infrastructure. For some of these devices, losing connectivity to them means that even very basic functions cease to work. A possible approach to this problem would be to analyze traffic between sensor-actuator pairs while online, build a model for each pair, and predict cloud server responses based on this model when the network loses internet connectivity. An important precursor to this approach is identifying sensor-actuator pairs in an IoT network, with no prior knowledge except for the hardware addresses of each IoT device - the association detection problem. This paper will discuss shortcomings of earlier approaches; describe a solution to the association detection problem using a modified Apriori algorithm, along with a method to create its input from network traffic; and finally, modify the solution to adapt to fluctuating network conditions. The final design accurately discovers sensor-actuator pairs using a simple approach with relatively low computational complexity, and with only minimal initial information. It will become an important first step towards providing a full offline reliability solution for IoT networks.