Blockchain and federated learning for secure and decentralized real-time IoT data analytics

Abstract

The rapid expansion of the Internet of Things (IoT) has led to unprecedented growth in real-time data generation, and this growth is raising issues concerning data security, privacy, and scalability. Conventional data processing mechanisms based on centralization are confronted with high latency, single points of failure, and vulnerability to cyberattacks. To address such issues, in this study, blockchain and Federated Learning (FL) are being used to implement an end-to-end secure and decentralized system for IoT data analytics in real time. Federated Learning enables IoT devices such as wearable health sensors, industrial sensors, and home automation devices to locally train AI models and transfer model updates rather than raw data and thereby ensure privacy and avoid communication overhead. To ensure security and trustfulness in model updates, a blockchain network using Hyperledger Fabric and Quorum is integrated with FL to avoid tampering and keep the process transparent with a decentralized ledger. Smart contracts are employed to authenticate and aggregate model updates so that only trusted devices can participate in training. Recurrent Neural Networks (RNN), Reinforcement Learning (RL), and Random Forest (RF) models are employed to enhance learning efficiency in the research. A database of 3,450 records is obtained from IoT sensors, and performance is quantified in terms of accuracy in the models, blockchain transaction speed, latency, energy usage, and bandwidth efficiency. The findings show that RNN is 97.86%, RL 93.4%, and RF 90.23%, confirming the success of the proposed system. The research highlights the potential of blockchain-FL integration for privacy-preserving training of AI in large-scale IoT applications, making it relevant to application areas such as healthcare, finance, and industrial automation.