Authors:
Richmond Osei
1
;
Habib Louafi
2
;
Malek Mouhoub
1
and
Zhongwen Zhu
3
Affiliations:
1
Department of Computer Science, University of Regina, Regina, SK, Canada
;
2
Department of Computer Science, New York Institute of Technology (Vancouver Campus), Canada
;
3
GAIA Montreal, Ericsson Canada, Montreal, Canada
Keyword(s):
Internet of Things IoT, Device Fingerprinting, Feature Extraction, Dimensionality Reduction, Machine Learning.
Abstract:
Internet of Things (IoT) usage is steadily becoming a way of life. IoT devices can be found in smart homes, factories, farming, etc. However, skyrocketing of IoT devices comes along with many security concerns due to their small and constrained build-up. For instance, a comprised IoT device in a network presents a vulnerability that can be exploited to attack the entire network. Since IoT devices are usually scattered over vast areas, Mobile Network Operators resort to analyzing the traffic generated by these devices to detect the identity (fingerprint) and nature of these devices (legitimate, faulty, or malicious). We propose an efficient solution to fingerprint IoT devices using known classifiers, alongside dimensionality reduction techniques, such as PCA and Autoencoder. The latter techniques extract the most relevant features required for accurate fingerprinting while reducing the amount of IoT data to process. To assess the performance of our proposed approach, we conducted seve
ral experiments on a real-world dataset from an IoT network. The results show that the Autoencoder for dimensionality reduction with a Decision Tree Algorithm reduces the number of features from 14 to 5 while keeping the prediction of the IoT devices fingerprints very high (97%).
(More)