Multilayer Perceptron Algorithms for Network Intrusion Detection on Portable Low Power Hardware

Abstract

Existing technology trends have led to an abundance of household items containing microprocessors all connected within a private network. Thus, network intrusion detection is essential for keeping these networks secure. However, network intrusion detection can be extremely taxing on battery operated devices. Thus, this work presents a cyberattack detection system based on a multilayer perceptron neural network algorithm. To show this system is capable of operating at low power, the algorithm was executed on two commercially available minicomputer systems including the Raspberry PI 3 and the Asus Tinkerboard. An accuracy, power, energy, and timing analysis was performed to study the tradeoffs necessary when executing these algorithms at low power. Our results show that these low power implementations are feasible, and a scan rate of more than 226,000 packets per second can be achieved from a system that requires approximately 5W to operate with greater than 99% accuracy.