AMIDS: An Adaptive Meta Learning Intrusion Detection System for Cross‐Variant Attack Adaption in IoT‐Based Smart Homes
International Journal of Network Management · 2026-10 · 36 (6)
DOI: 10.1002/nem.70059 ↗Kurz gefasst
Fachartikel in der Zeitschrift International Journal of Network Management, online erschienen am 09.11.2026. Verfasst von Waleed Hussain Ahmed Khan und 3 weiteren Autor:innen. Die Arbeit nennt 23 Literaturangaben.
Abstract (Originalsprache)
Smart homes built on Internet of things (IoT) devices are exposed to attacks that evolve rapidly, making it difficult for conventional intrusion detection systems (IDSs) to detect them. Signature‐based and machine learning detectors are fixed once training ends, require large volumes of labelled traffic, and must be retrained whenever a new attack appears, which is not practical for a residential gateway. This paper presents AMIDS, an adaptive meta‐learning intrusion detection system that pairs a base detector with a model agnostic meta learning (MAML) adaptation module, so that a previously unseen attack can be learned from a small number of labelled samples. The framework is evaluated on the CICIoMT2024 dataset, using all three of its published label sets under two setups. In the first setup, with a stratified data split in which all classes are used during training, MLP, CNN, LSTM, and SVM detectors and the proposed adaptive model are implemented, and their results are compared. In the second setup, one attack class is withheld from training and appears at test time only through a small support set. Thirteen such scenarios were run, each one a cross‐variant case in which the withheld class is a variant of an attack category the model has trained on under a different protocol or a different number of sources. On the same query records, the conventional detectors never assign the withheld label, and in three scenarios two or three of the four classify the withheld attack as benign traffic, while the adaptive model reaches an attack‐class F1 of 0.9722 or above from a single labelled sample in all 13 scenarios. Compared with a matched baseline that adapts the same network without meta‐training, meta‐training gives a clear gain in three scenarios, of up to 0.20 in F1 at one sample, and no material difference in the others.
Zitieren
APA
Khan, W. H. A., Abdulghafar, M., Nadeem, A., & Siddiqui, M. S. (2026). AMIDS: An Adaptive Meta Learning Intrusion Detection System for Cross‐Variant Attack Adaption in IoT‐Based Smart Homes. International Journal of Network Management, 36(6). https://doi.org/10.1002/nem.70059
Vancouver
Khan WHA, Abdulghafar M, Nadeem A, Siddiqui MS. AMIDS: An Adaptive Meta Learning Intrusion Detection System for Cross‐Variant Attack Adaption in IoT‐Based Smart Homes. International Journal of Network Management. 2026;36(6). doi:10.1002/nem.70059
BibTeX
@article{Khan2026,
author = {Khan, Waleed Hussain Ahmed and Abdulghafar, Mohammad and Nadeem, Adnan and Siddiqui, Muhammad Shoaib},
title = {AMIDS: An Adaptive Meta Learning Intrusion Detection System for Cross‐Variant Attack Adaption in IoT‐Based Smart Homes},
journal = {International Journal of Network Management},
year = {2026},
volume = {36},
number = {6},
doi = {10.1002/nem.70059}
}
EUVIMED