Beyond Signatures: Adaptive Machine Learning for IoT Threat Detection in Edge-to-Cloud Ecosystems
Journal of Sensors, IoT & Health Sciences · 31.03.2026 · 4 (1) · 28-56
DOI: 10.69996/jsihs.2026003 ↗Kurz gefasst
Fachartikel in der Zeitschrift Journal of Sensors, IoT & Health Sciences, online erschienen am 31.03.2026. Verfasst von Kritika Er.
Abstract (Originalsprache)
The attack surface has widened exponentially with the creation of Internet of Things (IoT) devices, which is estimated to grow to 30 billion in 2030, which makes conventional signature-based systems of intrusion detection outdated. Machine learning (ML) is now an impressive answer to anomalydetection, but the discipline is still divided with inconsistent datasets, methodological variation, and unexplored deployment limitations.The study presents a SLR of 50 peer-reviewed publications using PRISMA 2020 guidelines (from 2020-2025). The systemic review examines supervised, unsupervised, deep learning, and federated learning methods in the context of healthcare, smart home, industrial internet of things (IIoT), and smart city.It has been analyzed that the paradigm has shifted to privacy-preserving federated and edge-native frameworks (68% of studies, 2022-2025) instead of centralized deep learning frameworks (52% of studies, 2020-2021).Although 78.3% papers claim data accuracy above 95%, it turns out that only 17.4% of the articles test adversarial robustness and most of them use artificial datasets that are not likely applicable in diverse real encounters of the IoT in practice.Critical gaps have been revealed to be lack of data standardization, energy-efficiency evaluation, and scaling up deployment examination, whereas three paradigm-shifting fields are identified, such as federated learning, edge-native anomaly detection, and explainable AI. The research agenda suggested puts development of harmonized benchmarks, lightweight adversarially-robust models, and cross-domain generalization frameworks on its priority list.The SLR brings together the disjointed studies to a stage of maturity in the field of ML-based IoT anomaly detection and enhances a roadmap of finding ways to translate innovation in the laboratory to a production-ready security system.
Zitieren
APA
Er, K. (2026). Beyond Signatures: Adaptive Machine Learning for IoT Threat Detection in Edge-to-Cloud Ecosystems. Journal of Sensors, IoT & Health Sciences, 4(1), 28-56. https://doi.org/10.69996/jsihs.2026003
Vancouver
Er K. Beyond Signatures: Adaptive Machine Learning for IoT Threat Detection in Edge-to-Cloud Ecosystems. Journal of Sensors, IoT & Health Sciences. 2026;4(1):28-56. doi:10.69996/jsihs.2026003
BibTeX
@article{Er2026,
author = {Er, Kritika},
title = {Beyond Signatures: Adaptive Machine Learning for IoT Threat Detection in Edge-to-Cloud Ecosystems},
journal = {Journal of Sensors, IoT & Health Sciences},
year = {2026},
volume = {4},
number = {1},
pages = {28--56},
doi = {10.69996/jsihs.2026003}
}
EUVIMED