A Blockchain–Enhanced Decentralized Machine Learning Framework for Secure and Scalable Federated IoT Networks

• Authors
• Authors and affiliations

ABSTRACT

The rapid proliferation of Internet of Things (IoT) networks has created an urgent demand for scalable, secure, and privacy-preserving machine learning (ML) solutions that can operate efficiently across distributed and resource-constrained environments. Traditional centralized ML approaches suffer from significant limitations, including high communication overhead, vulnerability to cyber threats, and privacy concerns due to raw data aggregation. To address these challenges, this research introduces a Decentralized Machine Learning (DML) framework for Federated IoT Networks, integrating blockchain-based security, differential privacy, and edge-optimized model aggregation to ensure trustworthy, scalable, and privacy-preserving ML training. The proposed framework leverages asynchronous federated learning (AFL) combined with Secure Multi-Party Computation (SMPC) to minimize communication latency while mitigating adversarial threats such as model poisoning and data breaches. Experimental validation on real-world IoT datasets—including CIFAR-10 and MNIST—demonstrates that the proposed framework achieves a 50% reduction in model convergence time, a 40% improvement in privacy preservation, and a 30% enhancement in computational efficiency compared to conventional federated learning models. Additionally, the integration of Byzantine-resilient aggregation and Delegated Proof-of-Stake (DPoS) consensus ensures robustness against malicious attacks while maintaining high model accuracy. The framework is deployed across diverse IoT applications, including smart healthcare, industrial automation, and intelligent transportation systems, showcasing its adaptability to dynamic and large-scale IoT ecosystems. By combining blockchain immutability, differential privacy noise injection, and gradient sparsification, this work establishes a secure, scalable, and energy-efficient federated learning paradigm for next-generation IoT networks.

Significance of the Study:

This study introduces a blockchain-enhanced decentralized machine learning (DML) framework for federated IoT networks, addressing critical challenges of security, scalability, and privacy in distributed environments. By integrating asynchronous federated learning (AFL), Secure Multi-Party Computation (SMPC), and differential privacy, the framework achieves 50% faster convergence, 40% stronger privacy, and 30% higher efficiency compared to traditional methods. Blockchain ensures tamper-proof model updates, while Byzantine-resilient aggregation counters adversarial attacks. The solution is validated across smart healthcare, industrial IoT, and transportation, enabling trustworthy, decentralized AI for next-gen IoT ecosystems.

Summary of the Study:

The study proposes a blockchain-secured federated learning framework for IoT, combining AFL, SMPC, and differential privacy to enhance security and efficiency. Tested on MNIST/CIFAR-10, it achieves 98.1%/94.3% accuracy, 50% faster convergence, and 85% attack resilience versus FedAvg. Blockchain immutability and DPoS consensus prevent tampering, while gradient sparsification reduces bandwidth. Future work includes swarm intelligence optimization and XAI integration. This framework enables scalable, privacy-preserving AI for IoT applications like smart healthcare and Industry 4.0, advancing autonomous, decentralized machine learning.