Uncertainty-aware authentication model for IoT

Authors: Heydari, M., Mylonas, A., Katos, V., Balaguer-Ballester, E., Altaf, A. and Tafreshi, V.H.F.

Journal: Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

Volume: 11980 LNCS

Pages: 224-237

eISSN: 1611-3349

ISSN: 0302-9743

DOI: 10.1007/978-3-030-42048-2_15

Abstract:

Handling the process of authentication for the hundred million of computer embedded devices in Internet of Things (IoT) is not achievable without considering inherent IoT characteristics like scalability, heterogeneity, dependency and dynamism. In one hand, traditional and emerging access control models cannot handle indeterminate data access scenarios in IoT by applying deterministic access policies. On the other hand, moving towards resilient access control paradigms needs new attitudes and current manual risk analysis methods that rely on vulnerability calculations do not fit in IoT. This holds true as considering vulnerability as the key player in risk assessment is no longer efficient way to tackle with indeterminate access scenarios due to complicated dependency and scalability of IoT environment. Moreover, most of the IoT devices are not patchable so by discovering new vulnerabilities the vulnerable devices need to be replaced. Therefore, IoT needs agile, resilient and automatic authentication process. This work suggests a novel authentication method based on our previous work in which uncertainty was introduced as one of the neglected challenges in IoT. Uncertainty in authentication derived from incomplete information about incident happening upon authenticating an entity. Part of IoT characteristics makes such an uncertainty worse. Therefore, we have proposed an uncertainty-aware authentication model based on Attribute-Based Access Control (ABAC). Our prediction model is able to consider the uncertainty factor of mobile entities as well as fixed ones in authentication. In doing so, we have built our prediction model using boosting classifiers (AdaBoost and Gradient Boosting algorithms) besides voting classifier. We have compared the results with our previous work. Our designated model (AdaBoost) can achieve authentication performance with 86.54% accuracy.

Source: Scopus