A Lightweight Federated-Learning Intrusion Detection System for Agricultural IoT: Random-Forest-Guided Feature Selection with a Temporal Convolutional Classifier
International Journal of Advanced Computer Science and Applications · 2026 · 17 (10)
DOI: 10.14569/ijacsa.2026.0171086 ↗Kurz gefasst
Fachartikel in der Zeitschrift International Journal of Advanced Computer Science and Applications, 2026. Verfasst von Mohd Abdul Rahim Khan und 1 weiteren Autor:innen.
Abstract (Originalsprache)
Internet of Things (IoT) devices used in precision agriculture, such as soil and climate sensors, irrigation controllers, livestock trackers, and autonomous machinery, form a large, heterogeneous, and often poorly secured attack surface, referred to here as Agricultural IoT (Ag-IoT). Centralized intrusion detection is difficult to apply in this setting because raw traffic often cannot leave the farm gateway for privacy, bandwidth, and regulatory reasons, and because edge devices have limited computing and memory budgets. This study proposes a lightweight federated intrusion detection framework for Ag-IoT that decouples a one-time, centralized Random Forest (RF) feature-selection stage from repeated federated training of a compact Temporal Convolutional Network (TCN). TCN is trained with Federated Averaging (FedAvg) across ten simulated clients whose label distributions follow a Dirichlet distribution (α = 0.5), producing non-independent and identically distributed (non-IID) data. The RF reduces the UNSW-NB15 flow features to 20 high-importance attributes plus one RF attack-probability feature. On two hold-out protocols drawn from the official UNSW-NB15 training and testing partitions, the federated TCN reaches 95.84% accuracy (Matthews correlation coefficient, MCC = 0.904; area under the receiver operating characteristic curve, AUC = 0.988) and 97.72% accuracy (MCC = 0.954), respectively, with only 8,834 trainable parameters, a 0.034 MB model, and 0.034 ms per-sample inference latency, and it converges within three to five communication rounds. A per-client loss-spread diagnostic reveals an approximately 50-fold disparity between the best- and worst-served clients that global accuracy conceals. The study is limited to binary detection on one benchmark; same-split centralized and published baselines, alternative aggregators, and additional IoT datasets are identified as future work.
Zitieren
APA
Khan, M. A. R., & Naif, Y. H. (2026). A Lightweight Federated-Learning Intrusion Detection System for Agricultural IoT: Random-Forest-Guided Feature Selection with a Temporal Convolutional Classifier. International Journal of Advanced Computer Science and Applications, 17(10). https://doi.org/10.14569/ijacsa.2026.0171086
Vancouver
Khan MAR, Naif YH. A Lightweight Federated-Learning Intrusion Detection System for Agricultural IoT: Random-Forest-Guided Feature Selection with a Temporal Convolutional Classifier. International Journal of Advanced Computer Science and Applications. 2026;17(10). doi:10.14569/ijacsa.2026.0171086
BibTeX
@article{Khan2026,
author = {Khan, Mohd Abdul Rahim and Naif, Yasir Hashim},
title = {A Lightweight Federated-Learning Intrusion Detection System for Agricultural IoT: Random-Forest-Guided Feature Selection with a Temporal Convolutional Classifier},
journal = {International Journal of Advanced Computer Science and Applications},
year = {2026},
volume = {17},
number = {10},
doi = {10.14569/ijacsa.2026.0171086}
}
EUVIMED