Authors: Rizvi, Syed; Scanlon, Mark; McGibney, Jimmy and Sheppard, John

Publication Date: October 2023

Publication Name: 14th Annual IEEE Ubiquitous Computing, Electronics & Mobile Communication Conference (IEEE UEMCON)

Abstract:

Internet of Things (IoT) and edge computing devices have become integral to corporate and industrial systems. These devices are prime targets for attackers due to their constant availability and potential access to sensitive data. Handling substantial data quantities, these devices pose challenges in identifying relevant forensic evidence and investigating abnormal activities. Thus, accurate network intrusion detection is crucial in these resource-constrained environments. In addition, robust IoT forensic readiness strategies are vital for effective investigation. Unlike traditional computer forensic readiness, these strategies adapt to heterogeneous architectures. This paper evaluates an approach that directly trains and deploys AI models on resource-constrained devices, securing networks and categorizing significant traffic for later investigation. The approach identifies and records potential malicious attacks in real-time with minimal overhead, suitable for constrained environments. The experimentation employed the IoT-23 dataset. The outcome of the evaluation revealed that each of the included algorithms achieved a classification accuracy of over 99% on a representative resource-constrained device.

Download:

BibTeX Entry:

@inproceedings{rizvi2023AI-IDS-Resource-Constrained,
author={Rizvi, Syed and Scanlon, Mark and McGibney, Jimmy and Sheppard, John},
title="{An Evaluation of AI-Based Network Intrusion Detection in Resource-Constrained Environments}",
booktitle="{14th Annual IEEE Ubiquitous Computing, Electronics & Mobile Communication Conference (IEEE UEMCON)}",
address= "New York, USA",
publisher="IEEE",
year=2023,
month=10,
abstract={Internet of Things (IoT) and edge computing devices have become integral to corporate and industrial systems. These devices are prime targets for attackers due to their constant availability and potential access to sensitive data. Handling substantial data quantities, these devices pose challenges in identifying relevant forensic evidence and investigating abnormal activities. Thus, accurate network intrusion detection is crucial in these resource-constrained environments. In addition, robust IoT forensic readiness strategies are vital for effective investigation. Unlike traditional computer forensic readiness, these strategies adapt to heterogeneous architectures. This paper evaluates an approach that directly trains and deploys AI models on resource-constrained devices, securing networks and categorizing significant traffic for later investigation. The approach identifies and records potential malicious attacks in real-time with minimal overhead, suitable for constrained environments. The experimentation employed the IoT-23 dataset. The outcome of the evaluation revealed that each of the included algorithms achieved a classification accuracy of over 99% on a representative resource-constrained device.}
}

---