SecureDyn-FL: A Robust Privacy-Preserving Federated Learning Framework for Intrusion Detection in IoT Networks

TL;DR

SecureDyn-FL tackles privacy-preserving intrusion detection in heterogeneous IoT via a federated learning framework that combines dynamic temporal gradient auditing, lightweight encryption-based aggregation, and a dual-objective personalized learning strategy. It introduces a GMM+MD auditing mechanism to detect stealthy poisoning, a post-Cramer transformed additive ElGamal encryption for secure gradient aggregation, and logit-adjusted personalization to mitigate non-IID data effects. Empirical evaluation on N-BaIoT and TON_IoT shows high accuracy (≈99%+) and strong privacy protection against gradient inversion and membership inference, outperforming state-of-the-art FL-IDS defenses across IID and non-IID settings, even with substantial adversarial participation. The work demonstrates practical viability for scalable, secure, and robust FL-based IDS in real-world IoT deployments, with clear avenues for future enhancements including advanced neural architectures and blockchain-enabled auditing.

Abstract

The rapid proliferation of Internet of Things (IoT) devices across domains such as smart homes, industrial control systems, and healthcare networks has significantly expanded the attack surface for cyber threats, including botnet-driven distributed denial-of-service (DDoS), malware injection, and data exfiltration. Conventional intrusion detection systems (IDS) face critical challenges like privacy, scalability, and robustness when applied in such heterogeneous IoT environments. To address these issues, we propose SecureDyn-FL, a comprehensive and robust privacy-preserving federated learning (FL) framework tailored for intrusion detection in IoT networks. SecureDyn-FL is designed to simultaneously address multiple security dimensions in FL-based IDS: (1) poisoning detection through dynamic temporal gradient auditing, (2) privacy protection against inference and eavesdropping attacks through secure aggregation, and (3) adaptation to heterogeneous non-IID data via personalized learning. The framework introduces three core contributions: (i) a dynamic temporal gradient auditing mechanism that leverages Gaussian mixture models (GMMs) and Mahalanobis distance (MD) to detect stealthy and adaptive poisoning attacks, (ii) an optimized privacy-preserving aggregation scheme based on transformed additive ElGamal encryption with adaptive pruning and quantization for secure and efficient communication, and (iii) a dual-objective personalized learning strategy that improves model adaptation under non-IID data using logit-adjusted loss. Extensive experiments on the N-BaIoT dataset under both IID and non-IID settings, including scenarios with up to 50% adversarial clients, demonstrate that SecureDyn-FL consistently outperforms state-of-the-art FL-based IDS defenses.

Figures (11)

• Figure 1: Thread Model: Federated Learning-based Intrusion Detection System (FL-IDS) in IoT networks with model inference and poisoning attacks.
• Figure 2: Flowchart of the SecureDyn-FL framework, including device initialization, local model training on clients, security measures (e.g., poisoned device detection and central auditing), optimization (via weight pruning and quantization), privacy-preserving aggregation at the server, and performance evaluation before deploying the global model.
• Figure 3: System model of proposed SecureDyn-FL framework. The architecture integrates multiple defense mechanisms, including quantization, mean-based parameter clipping, and dynamic unstructured model pruning, followed by encrypted local training. The FL model is decoupled into a shared feature extractor and dual classifiers (global and personalized) enabling multi-objective optimization. A trusted auditor monitors the training pipeline to detect adversarial threats such as data/model poisoning and eavesdropping. Secure aggregation ensures the integrity and privacy of the global model.
• Figure 4: Analyzing malicious alarms on IID data
• Figure 5: Analyzing malicious alarms on non-IID data

Theorems & Definitions (8)

• Lemma 1
• Lemma 2
• Lemma 3
• Lemma 4
• Theorem 1: Robustness against malicious gradients via auditability - IID
• Theorem 2
• proof
• Theorem 3: Robustness against malicious gradients via auditability - non-IID