Adaptive Lightweight Federated Learning With Aggregation-Only CKKS for Privacy-Preserving IoT Intrusion Detection

Mahdi Ajdani
Additional contact information
Mahdi Ajdani: Islamic Azad University of Qeshm, Iran

International Journal of Information Security and Privacy (IJISP), 2026, vol. 20, issue 1, 1-19

Abstract: Federated learning (FL) enables collaborative training without sharing raw data, but standard FL exposes client updates and burdens resource-constrained IoT devices. The authors propose AdaptiveCKKS, an FL framework combining aggregation-only CKKS encryption with index-free block sparsification and stochastic quantization. A lightweight controller adaptively selects compression ratio and quantization per device/round based on on-device calibration of bandwidth, CPU, and encryption cost, while CKKS contexts are fixed at enrollment. The server performs ciphertext-only additions, decrypting only the aggregate each round. On BoT-IoT and ToN-IoT datasets, AdaptiveCKKS improves accuracy by 3.2–3.8% over FL and fixed-HE, reduces per-round communication by ~45% and average power by ~39%, and increases resistance to membership inference and gradient inversion attacks. Results are averaged over 10 runs with 95% confidence intervals, and all artifacts are released for reproducibility.

Date: 2026
References: Add references at CitEc
Citations:

Downloads: (external link)
https://services.igi-global.com/resolvedoi/resolve.aspx?doi=10.4018/IJISP.402007 (application/pdf)

Related works:
This item may be available elsewhere in EconPapers: Search for items with the same title.

Export reference: BibTeX RIS (EndNote, ProCite, RefMan) HTML/Text

Persistent link: https://EconPapers.repec.org/RePEc:igg:jisp00:v:20:y:2026:i:1:p:1-19

Access Statistics for this article

International Journal of Information Security and Privacy (IJISP) is currently edited by Yassine Maleh

More articles in International Journal of Information Security and Privacy (IJISP) from IGI Global Scientific Publishing
Bibliographic data for series maintained by Journal Editor ().