Tantangan dan Tren Masa Depan Keamanan Federated Learning pada Ekosistem IoT Tinjau Pustaka Sistematis

Abstrak

The rapid growth of the Internet of Things (IoT) ecosystem has generated a massive surge in data, driving the utilization of Federated Learning (FL) as a decentralized model training solution that guarantees user privacy. Despite its promise, the implementation of FL in IoT infrastructures is frequently hindered by cybersecurity vulnerabilities and hardware computational constraints. Therefore, this paper presents a Systematic Literature Review (SLR) guided by the PRISMA 2020 standards to analyze the architectures, security protocols, and future projections of FL in IoT environments. Through the extraction and synthesis of 25 selected primary literatures indexed in the Scopus database, the findings indicate that Deep Learning models-particularly CNN and LSTM-are the most dominant approaches. Furthermore, the combination of Homomorphic Encryption and Differential Privacy is established as the primary benchmark in mitigating model inversion attacks. In terms of constraints, the high communication overhead during gradient exchange on edge devices remains the largest operational hurdle. Ultimately, this study concludes that the integration of FL with Serverless infrastructure and Blockchain technology represents a highly potential future innovation to eliminate centralized failure risks while realizing a secure and autonomous IoT network.

Identifier

• DOI: https://doi.org/10.33365/jtikom.v7i1.1953
• OpenAlex: https://openalex.org/W7172541791