SecureDyn-FL Boosts IoT Intrusion Detection with Privacy-Preserving Federated Learning

Global: SecureDyn-FL Boosts IoT Intrusion Detection with Privacy-Preserving Federated Learning

Background

A team of researchers has introduced SecureDyn-FL, a federated learning framework designed to improve intrusion detection across diverse Internet of Things (IoT) environments. The approach targets the growing attack surface presented by smart homes, industrial control systems, and healthcare networks, where traditional intrusion detection systems (IDS) often struggle with privacy, scalability, and robustness.

Framework Overview

SecureDyn-FL addresses three critical security dimensions in federated IDS: detection of poisoning attacks, protection against inference and eavesdropping, and adaptation to heterogeneous non‑IID data. By integrating these capabilities, the framework aims to provide a comprehensive solution for real‑world IoT deployments.

Dynamic Gradient Auditing

The system incorporates a dynamic temporal gradient auditing mechanism that employs Gaussian mixture models and Mahalanobis distance calculations to identify stealthy and adaptive poisoning attempts. This continuous monitoring of gradient patterns enables early detection of malicious client behavior without requiring centralized data access.

Secure Aggregation

To safeguard model updates, SecureDyn-FL utilizes a transformed additive ElGamal encryption scheme enhanced with adaptive pruning and quantization. This privacy‑preserving aggregation reduces communication overhead while preventing inference attacks and eavesdropping on the federated learning process.

Personalized Learning

A dual‑objective personalized learning strategy is introduced to handle non‑IID data distributions common in IoT networks. By applying a logit‑adjusted loss function, the framework improves model convergence for individual clients, fostering better detection performance across heterogeneous devices.

Experimental Validation

Extensive experiments on the N‑BaIoT dataset, conducted under both IID and non‑IID conditions and with up to 50% adversarial clients, show that SecureDyn-FL consistently outperforms existing federated IDS defenses. Metrics indicate superior detection accuracy and resilience to poisoning compared with state‑of‑the‑art baselines.

Implications and Future Directions

The results suggest that privacy‑preserving federated learning can effectively strengthen IoT security without compromising data confidentiality. Researchers anticipate further refinement of the auditing and aggregation components, as well as broader testing on additional IoT datasets to validate scalability.

This report is based on information from arXiv, licensed under Academic Preprint / Open Access. Based on the abstract of the research paper. Full text available via ArXiv.