Machine Learning for IoT based networks intrusion detection: a comparative study

Abstract

Internet of Things (IoT) refers to technologies that enable the connection of objects to the internet and also collect data with minimal human intervention. Nowadays IoT is applied to all areas, particularly in logistics, industry, and health. These new IoT applications expand the functionality of smart objects and also introduce security vulnerabilities. In this context, the development of intrusion detection systems (IDS) adopted to IoT networks has become an important field of research. Machine learning (ML) and deep learning (DL) approaches are solutions used by many researchers in IDS. In this work, we propose a state of the art on IoT network intrusion detection using ML techniques during the last few years. We aim to detect the most used and efficient machine learning techniques. To support and confirm the proposed state-of-the-art results, we carried out an experiment to compare the machine learning techniques extracted from our analysis on the same dataset. we also analyzed the performance and execution time of two feature selection techniques and a feature extraction technique for intrusion detection in the case of binary classification, and multi-classification (5 classes: Normal, DOS, Probe, U2R, and R2L), The experimental results reveal that the Decision Tree with Fisher score gave the best performance with an accuracy of 99.26% and a minimum prediction time of 0.4 seconds.