Resource-constrained Internet of Things (IoT) devices are increasingly deployed in critical domains but remain vulnerable to stealthy attacks that can bypass conventional defenses. At the same time, privacy constraints limit centralized data collection and processing, complicating anomaly detection. This systematic review surveys methods for privacy-preserving anomaly detection in resource-constrained IoT and introduces a five-dimension taxonomy covering deployment paradigms, resource constraints, real-time requirements, protection techniques, and communication constraints. We review how the literature measures and reports resource and privacy costs and identify three major gaps: (1) a shortage of co-designed detector-plus-privacy solutions tailored to constrained hardware, (2) inconsistent reporting of resource and privacy trade-offs, and (3) limited robustness against adaptive attackers and realistic deployment noise. We conclude with actionable recommendations and a prioritized research roadmap. Furthermore, the multi-dimensional taxonomy we introduce provides a structured framework to guide design choices and systematically improve the comparability, deployability, and overall trustworthiness of anomaly detection systems for constrained IoT.
Neagu et al. (2026) studied this question.