Federated IoT Intrusion Detection Dataset - Privacy-Preserving Security

Dataset Overview

This pioneering dataset published in Nature Scientific Reports in January 2026 addresses the critical challenge of IoT security in privacy-sensitive environments. It enables research into federated learning approaches where intrusion detection models are trained across distributed IoT networks without centralizing raw data.

Distributed Network Architecture

The dataset simulates multiple independent IoT deployments:

• Number of Nodes: 10+ independent IoT network sites
• Device Diversity: Each site contains different device types and manufacturers
• Network Topologies: Varied architectures (star, mesh, hierarchical)
• Geographic Distribution: Simulated sites in different regions with varying threat landscapes

Privacy-Preserving Data Structure

Local Network Traffic

Each site provides:

• Flow-Based Features: Aggregated traffic statistics without raw packets
• Attack Labels: Local intrusion detection annotations
• Site Metadata: Anonymous identifiers and configuration parameters

Federated Learning Metrics

Performance measures for distributed training:

• Communication Rounds: Number of model update exchanges
• Model Convergence: Accuracy improvement across federation rounds
• Data Heterogeneity: Statistical divergence between sites
• Privacy Budget: Differential privacy parameters (epsilon, delta)

Attack Coverage

Each site contains varying proportions of attack types:

• DDoS and DoS attacks
• Port scanning and reconnaissance
• MITM attacks
• Botnet command-and-control traffic
• Data exfiltration attempts

The heterogeneous attack distribution tests federated models' ability to generalize across diverse threat environments.

Research Contributions

Dataset-Centric Evaluation

The publication emphasizes evaluating federated learning algorithms based on data characteristics rather than just model architectures, providing insights into when federated approaches outperform centralized training.

Benchmark Results

Baseline performance for multiple federated learning algorithms:

• FedAvg (Federated Averaging)
• FedProx (Federated Proximal)
• SCAFFOLD (Stochastic Controlled Averaging)
• FedOpt (Federated Optimization)

Practical Applications

• Multi-Organization Security: Collaborative threat detection across competing organizations without sharing sensitive data
• GDPR Compliance: Privacy-preserving security analytics meeting regulatory requirements
• Edge-Cloud Hybrid: Distributed learning between edge devices and cloud infrastructure
• Continuous Adaptation: Models improving over time through federated updates without data movement

Academic Significance

Published in Nature with rigorous peer review, this dataset advances both IoT security and privacy-preserving machine learning fields. It provides reproducible benchmarks for evaluating federated learning in real-world IoT security scenarios.