Towards sustainable IoT: A digital signature‐enhanced federated learning approach

Abstract

Federated Learning (FL) is emerging as a premier paradigm for privacy‐preserved Machine Learning (ML), enabling devices to train models without central data pooling collaboratively. In the contemporary Internet of Things (IoT) landscape, characterized by escalating energy consumption and associated carbon footprint, FL is recognized not merely for its privacy features. Intrinsic to decentralized architectures such as FL, secure communication is based on digital signatures to guarantee integrity. This is particularly evident in sensitive sectors such as the Internet of Vehicles (IoV), banking, and healthcare. Integrating FL becomes imperative and intricate as these sectors are intertwined with the IoT fabric. Our study unveils “Secure Federated Learning Framework (SecFL),” a pioneering decentralized framework combining FL and sustainable computing. SecFL offers defences against adversarial attacks such as data poisoning and label flipping. Utilizing the Rivest‐Shamir‐Adleman (RSA) asymmetric encryption algorithm for securing digital communications and transactions, combined with ElGamal encryption and a private Ethereum blockchain, ensures enhanced client‐specific security. Our research emphasizes the formal modeling of adversarial dynamics using High‐Level Petri nets (HLPN) within the FL‐IoT ecosystem, balancing system dynamics and energy conservation. Our model consistently outperforms contemporary solutions in accuracy and time efficiency after validation. As IoT burgeons into domains like environmental monitoring, smart cities, and energy grids, the SecFL framework, fostering FL, optimizes energy utilization and bolsters resource efficiency. In our comparative analysis, the Elliptic Curve Digital Signature Algorithm (ECDSA) algorithm demonstrates superior transaction latency and verification time compared to RSA and Elliptic Curve Cryptography (ECC).