Cyber Attack Detection System for Internet of Things Using Machine Learning: A Review

Abstract

With the explosive growth of the Internet of Things (IoT), intrusion detection systems have been challenged with new and complex cybersecurity problems that they are ill equipped to handle. Therefore, in the recent work ML-methods were highlight as one of the appropriate solution that can be used for assuring intelligent and adaptive (real-time) perceiving an online cyberattacks attacking IoT environments. To the best of our knowledge, no specific systematic review paper has been published on ML-based intrusion detection techniques in IoT systems; thus, this study aims to provide an overview and comparison of such methods with respect to underlying methodologies, datasets utilized for experimentation purpose, performance metrics that have been evaluated upon and deployment challenges faced out. Although the available works mainly focus on enhancing detection performance, important characteristics of scalable estimation and robustness, light computation load, interpretable model design as well as adaptability to diverse IoT environments still need to be further investigated. Solutions to these limitations are important in realizing practical and effective intrusion detection solutions for next generation of IoT networks.For the period 2020–2025, the latest developments are reviewed, and a unified classification is proposed for ML based IoT intrusion detection technologies, based on learning models and operational contexts. The contributions included in this study are a systematic comparative analysis of ML models also IoT security datasets, providing insights into future research directions towards hybrid, interpretable, and privacy-preserving frameworks in the field of ML for cyber IoT security, and identifying important open issues related to resource limitations and model interpretability.