An Intelligent Intrusion Detection System for IoT Networks Using Machine Learning Algorithms: A Comprehensive Review

Internet of Things (IoT) is a fast developing phenomenon with billions of devices used in houses, industries, medical services, and transport. Although such connectivity enhances automation and intelligence, it also raises vulnerability to cyber threats including DDoS attacks, botnets, spoofing and malware. Conventional signature-based Intrusion Detection Systems (IDS) can hardly identify new and emerging attacks, and hence the Machine Learning (ML)-based IDS is a better solution. This review identifies different ML methods such as supervised, unsupervised, deep learning, ensemble models, and hybrid models and considers the popular datasets, such as Bot-IoT, TON_IoT, UNSW-NB15 and CICIDS2017. It also addresses the most important key performance metrics including accuracy, precision, recall, F1-score, and detection latency. Severe limitations are the insufficient resources of devices, asymmetric datasets, scaling, the possibility of detecting threats at a zero-day, and privacy. New solutions like federated learning, edge-based IDS, graph neural networks, block chain and Explainable AI have a lot of potential towards improving IoT security. All in all, IDS architecture based on ML has a major role in improving the resilience and reliability of future IoT systems.