Intrusion Detection Systems in IoT Based on Machine Learning: A state of the art

Abstract

Nowadays, connected devices are progressively integrating our lifestyle, known as the Internet of Things (IoT). There are several domains where these IoT can be found, including industry, smart cities, energy, healthcare, smart homes, and so on. These devices bring with them a host of security threats, due to their many weaknesses such as limited resources and the fact that it is often out of date. This is why researchers have progressively turned to innovative techniques to face these threats, such as Intrusion Detection Systems (IDS). In addition, IDS can be enhanced through machine learning (ML), which improves detection accuracy, adapts to new and evolving threats, identifies unknown threats, reduces false positives, and more. In this study, we conduct a review of the recent articles focusing on the integration of IDS and ML. Each method is evaluated based on attributes such as feature selection method, algorithms used, detection and false positive rates, security threats, and so on. This involves presenting a critical analysis of each approach. Furthermore, we present the most prevalent datasets containing deliberate attacks on test IoT networks. In addition, we cover the anomaly detection-based IDS method in depth, focusing on papers that use ML. Subsequently, we engage in a discussion by comparing all methods and identifying open re-search questions. As a result, summarized tables are provided to highlight each model.