A Deep Learning Approach for Classifying Network Connected IoT Devices Using Communication Traffic Characteristics

The Internet of Things can be considered a technological revolution and has successfully merged the physical world with the digital world. However, heterogeneous IoT devices with different functionalities impose new security challenges in cyberspace, including node forgery, unauthorized access to data and cyberattacks. It is essential to identify network-connected devices accurately and robustly, as well as their communication behaviours, to improve network security. Whilst necessary for communication, traditional identifiers using internet protocol /medium access control addresses have some constraints as device identifiers due to vulnerabilities against different attacks. To mitigate these issues, a deep learning-based device fingerprinting model has been proposed using these two features for the classification task, with 100 consecutive packets' information utilized to generate fingerprints as graphs. The proposed device fingerprinting model demonstrates over 99% and 95% precisions in distinguishing between known and unknown traffic traces and in identifying IoT and non-IoT traffic traces, respectively. 98.49% precision has also been demonstrated on an individual device classification task. These results are significant as the model can be utilized to effectively secure a resource-constrained IoT network, which despite its rapid growth of usage, is more prone to attack, partly due to its dependence on traditional explicit identification methods.