Random Neural Network for Lightweight Attack Detection in the IoT

Author: Filus K.; Domańska J.; Gelenbe E.
Conference Name: MASCOTS 2020: Modelling, Analysis, and Simulation of Computer and Telecommunication Systems
Year of Publication: 2021
Status: Published
DOI: https://doi.org/10.1007/978-3-030-68110-4_5
Publisher: Springer International Publishing
URL: https://link.springer.com/chapter/10.1007/978-3-030-68110-4_5
PDF File: Random Neural Network for Lightweight Attack Detection in the IoT.pdf (336.06 KB)
Abstract: Cyber-attack detection has become a basic component of all information processing systems, and once an attack is detected it may be possible to block or mitigate its effects. This paper addresses the use of a learning recurrent Random Neural Network (RNN) to build a lightweight detector for certain types of Botnet attacks on IoT systems. Its low computational cost based on a small 12-neuron recurrent architecture makes it particularly attractive for edge devices. The RNN can be trained off-line using a fast simplified gradient descent algorithm, and we show that it can lead to high detection rates of the order of 96%, with false alarm rates of a few percent.