AI-Driven Intrusion Detection Systems for Autonomous Internet of things (A-IoT): Trends, Architectures and Evaluation Practices

Abstract

This paper is a systematic review of the state of the art of Intrusion Detection Systems proposed for internet of things scenarios. The emphasis was placed on understanding research trends and methodologies and also on understanding the impact of dataset diversity on intrusion detection. Using a systematic review protocol, 180 peer-reviewed studies published between 2010 and 2025 were analyzed. The paper utilizes a literature review methodology to investigate recent published articles related to IDS. The results indicate a steady rise in scientific interest in the use of IoT-based intrusion detection systems since 2018, and a resurgence in growth since 2022 due to the widespread use of technology and developments in artificial intelligence-based security systems The findings confirm that IDS models trained on diverse, representative datasets consistently outperform those evaluated on homogeneous or simulated datasets. The analysis also proves that simulation has dominated evaluation techniques but has been gradually replaced by experimental validation. Moreover, this analysis also highlights that a single IDS architecture is not capable of providing a holistic security solution for various IoT setups. The review identifies key research gaps in dataset standardization, real-world validation, and autonomous IDS adaptability, and provides guidance for future IDS design in autonomous IoT environments The decentralized, centralized, and hybrid architectures of IDS have their individual advantages and challenges in terms of scalability, latency, energy, and complexity of implementation. Finally, this analysis has shown that testing and security-based research in IoT setups must focus on adaptive and diverse benchmarking mechanisms.