Feasible design or validated practice? Unpacking the operational readiness of federated learning in industrial IoT
Federated learning (FL) is promoted as the way to build intrusion detection for the industrial Internet of Things (IIoT) without pooling sensitive plant traffic. This review asks whether published evidence shows detection that survives realistic distributed conditions rather than benchmark accuracy alone. Journal articles issued from 2022 to 2026 were retrieved from the OpenAIRE Graph and Crossref. A bibliometric map of 114 articles was combined with an abstract-level evidence map of 62 core studies, technical appraisal of six full texts, and a structured synthesis of 28 matched contrasts against FedAvg. Output grew from 5 articles in 2022 to 45 in the partial year 2026, resting on a few public captures. Among 50 core abstracts, 26% mentioned non-IID data, 6% a test on a new device, site, domain, or period, 4% a macro or per-class metric, 4% differential privacy parameters, 2% a hardware testbed, and none a threat model. On in-distribution test splits, proximal and normalized aggregation gained a median of 1.0 F1 point over FedAvg (range −0.9 to 3.0; 23 settings, 16 from one single-run study), whereas on the only chronological hold-out FedAvg outperformed FedProx by 1.9 points and FedAdam by 3.4. Shifts in heterogeneity, time, or environment cost 2 to 30 points. Privacy rested on data locality, and costs were rarely measured on hardware. No pooled estimate was defensible. The evidence supports federated intrusion detection for IIoT as a feasible design, not yet a validated deployment practice.
Authors
- Rafi Muhammad Nur
Publication Details
- Journal
- Journal of Engineering and High Technology
- Published
- 2026-11-15
- DOI
- https://doi.org/10.55942/jeht.v2i2.2368
- Primary Topic
- Privacy-Preserving Technologies in Data
- Type
- article
- Field-Weighted Citation Impact
- 0.00