An Improved IoT Device Fingerprinting via Enhanced Feature Selection

Abstract

In the context of the Internet of Things (IoT), device fingerprinting involves identifying or classifying devices based on their unique intrinsic behavioural properties found in the network traffic data, which can be vast in amount and contain redundant information. Therefore, feature selection from the available network data is crucial for device fingerprinting. Feature sets selected by the existing fingerprinting methods are domain-adapted, specific to the datasets in which the methods are tested. They do not perform well in cross-domain scenarios when tested against other datasets. In this paper, we propose a novel feature set for an improved device fingerprinting method, which performs better than the existing methods. Genetic algorithm and voting mechanisms are utilised to obtain the feature set, and the stratified sampling method is used to ensure unbiased data for training ML models. We compare the performance of the proposed feature set to various existing methods using three benchmark IoT network traffic datasets. Our results show that the new feature set successfully identified IoT devices with 98.4% accuracy under a testing time of 0.02 seconds. Show more