Intrusion Detection via MLP Neural Network Using an Arduino Embedded System

Abstract

Real-time intrusion detection using low-power devices is one of the main challenges for the Internet of Things (IoT) research community. Currently, many Intrusion Detection Systems tackle this task using Artificial Neural Networks (ANNs) and other machine learning techniques. However, some of these methods are computationally costly, which makes them unfeasible in an IoT scenario. To address this, we train a Multilayer Perceptron (MLP) using NLS-KDD for Weka, a modified version of the NSL-KDD dataset containing less features, resulting in a perceptron neural network with a small number of artificial neurons. As a result, we evaluated the MLP networks using metrics such as accuracy, precision, and coverage, as well as classifier performance running on Arduino via time measurements (microseconds).