Optimization-driven machine learning intrusion detection systems for resource-constrained internet of things challenges: a systematic review and future directions

Abstract

The explosive growth of the Internet of Things (IoT) across smart homes, healthcare, industrial systems, and critical infrastructure has dramatically enlarged the cyberattack surface, rendering traditional security mechanisms ineffective against evolving threats in highly heterogeneous and resource-constrained environments. This article presents a comprehensive and systematic review of machine learning (ML) and deep learning (DL) based intrusion detection systems (IDSs) tailored for IoT, with a particular focus on optimization algorithms for feature selection (FS), hyperparameter tuning, and model lightweighting. We critically analyze state-of-the-art methodologies, benchmark datasets, deployment strategies, and performance trade-offs, while uncovering previously hidden correlations between optimization techniques and ML/DL models that significantly enhance detection accuracy, interpretability, energy efficiency, and real-time feasibility on edge devices. The effectiveness of these approaches is systematically evaluated across diverse IoT applications, including real-time anomaly detection, lightweight edge deployment, and encrypted traffic analysis. Key challenges in integrating advanced ML/DL into constrained IoT environments are thoroughly examined, and a forward-looking research framework is proposed to overcome current limitations and drive the development of the next generation of intelligent, adaptive, and resilient IoT-IDS solutions.