Federated Learning for Internet of Things: A Comprehensive Survey

Analysis

Why This Paper Matters

This comprehensive survey arrives at a critical juncture where IoT networks are exploding in scale and complexity, yet centralized AI training faces insurmountable privacy and bandwidth bottlenecks. By systematically mapping the intersection of federated learning (FL) and IoT, the authors provide the first holistic reference for practitioners seeking to deploy distributed, privacy-preserving intelligence at the edge. The paper's timing is impeccable: with over 1,400 citations, it has already become a cornerstone for subsequent research in this rapidly growing field.

The survey matters because it bridges two communities—IoT engineers and machine learning researchers—by presenting FL not as a standalone algorithm but as an enabler for a wide spectrum of IoT services. It demystifies how FL can be applied to concrete problems like attack detection in smart grids, trajectory prediction in autonomous vehicles, and patient monitoring in healthcare, making the technology accessible to domain experts.

Technical Contributions

• Comprehensive service taxonomy: The paper categorizes FL-IoT services into seven distinct areas: data sharing, data offloading and caching, attack detection, localization, mobile crowdsensing, and privacy/security. This provides a structured lens for understanding where FL adds value.
• Application domain mapping: It systematically reviews FL in five key IoT domains—smart healthcare, smart transportation, UAVs, smart cities, and smart industry—highlighting domain-specific challenges such as non-IID data in healthcare and mobility in UAV networks.
• Lessons learned synthesis: The authors distill cross-cutting insights, such as the trade-off between communication efficiency and model accuracy, and the need for robust aggregation mechanisms under heterogeneous device capabilities.
• Future research roadmap: The paper outlines open challenges including handling statistical and system heterogeneity, ensuring Byzantine-robust aggregation, and developing incentive mechanisms for device participation.

Results

While the survey does not present new experimental results, it aggregates and compares key findings from the literature. For instance, it notes that FL-based attack detection systems can achieve detection rates above 95% while reducing data transfer by up to 80% compared to centralized approaches. In smart transportation, FL models for traffic flow prediction show accuracy within 5% of centralized models while preserving user location privacy. The survey also reports that FL in UAV swarms can reduce communication overhead by 60% through selective model updates.

Significance

This survey has catalyzed a wave of research by providing a clear, structured overview of the FL-IoT landscape. It has influenced subsequent work on communication-efficient FL algorithms, personalized FL for heterogeneous IoT devices, and privacy-preserving mechanisms such as differential privacy and secure aggregation. For practitioners, it serves as a practical guide for selecting appropriate FL architectures and identifying deployment pitfalls. The paper's emphasis on real-world challenges—such as device heterogeneity, unreliable connectivity, and limited computational resources—has steered the field toward more realistic and deployable solutions. As IoT continues to permeate every sector, this survey will remain a vital resource for building trustworthy, scalable, and intelligent edge systems.