EUVIMEDEUVIMED entdecken
Crossref

RedPO-BRNNet: Federated Anomaly Detection with Differential Privacy and Secure Aggregation for Edge IoT

Rolly R Tang

Informatica · 16.12.2025 · 49 (35)

DOI: 10.31449/inf.v49i35.11175 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Informatica, online erschienen am 16.12.2025. Verfasst von Rolly R Tang. Die Arbeit wurde bisher 2-mal zitiert (Crossref).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

In the context of edge computing (EC), there are distributed Internet of Things (IoT) devices that generate huge volumes of sensitive data, which necessitates privacy-preserving and efficient model training. Federated learning (FL) enables the ability to collaboratively train models without sharing raw data, albeit at the risk of sharing data values through indirect means and in an uncontrolled manner that continues to expose privacy risks like gradient inference attacks and data reconstruction methods. To propose a secure aggregation framework that combines FL with differential privacy (DP) at the 'noise' level of 1.e−4 and a red panda optimizer fused bidirectional recurrent neural network (RedPO-BRNNet), that improves predictive performance, convergence efficiency, and robustness in an EC environment. This framework is implemented using Python and utilizes local datasets of 2,501 time-series recordings gathered from 10 distributed edge devices, which feature blood pressure, activity level, body temperature, heart rate, anomaly scores, and a binary classification label. The data are pre-processed using Z-score normalization to standardize scales for the features across devices. Each edge device trains a local RedPO-BRNNet model, applies DP noise to the model parameters and then contributes updates that preserve privacy to the global aggregation using FedAvg. RedPO-BRNNet is tested against LSTM, Bi-LSTM, and baseline BRNN models during five training epochs. The proposed model achieves 94.97% anomaly detection accuracy, 2.14% privacy leakage, 57.82 ms client latency, and 78.96% model convergence efficiency, with improvements statistically validated using paired t-tests (p < 0.05). These results show that RedPO-BRNNet effectively protects privacy while enabling efficient and scalable edge intelligence, making it a dependable solution for safe, multi-device IoT deployments.

Zitieren

APA

Tang, R. R. (2025). RedPO-BRNNet: Federated Anomaly Detection with Differential Privacy and Secure Aggregation for Edge IoT. Informatica, 49(35). https://doi.org/10.31449/inf.v49i35.11175

Vancouver

Tang RR. RedPO-BRNNet: Federated Anomaly Detection with Differential Privacy and Secure Aggregation for Edge IoT. Informatica. 2025;49(35). doi:10.31449/inf.v49i35.11175

BibTeX
@article{Tang2025,
  author = {Tang, Rolly R},
  title = {RedPO-BRNNet: Federated Anomaly Detection with Differential Privacy and Secure Aggregation for Edge IoT},
  journal = {Informatica},
  year = {2025},
  volume = {49},
  number = {35},
  doi = {10.31449/inf.v49i35.11175}
}
Kostenlos starten →