Securing Internet-of-Things Systems Through Implicit and Explicit Reputation Models

Internet-of-Things (IoT) systems are usually composed of thousands of different components among hardware devices and different software modules. In order to address the design of these complex systems, different abstraction layers are usually defined. However, as these layers are isolated, high-level components always have uncertainty about the nature of the low-level components they relate with. In particular, as low-level component identities are not known by user applications, and current IoT systems are vulnerable to the injection of new components and to the modification of the behavior of existing ones (adequate security solutions at the network level for these problems have not been found yet), the reliability of the received data is generally compromised. In this context, new mechanisms are required to avoid the interactions or directly remove the malicious components relying on high-level information. This paper describes a statistical framework to discover IoT components with malicious behaviors, using a hybrid reputation model. On the one hand, an implicit reputation definition is employed, based on the observations made by a certain IoT component and other modules it relies on. On the other hand, an explicit reputation model considers a scheme of recommendations and negative grades. The proposed solution is evaluated in a simulation scenario by using the NS3 simulator, in order to perform an experimental validation.