Lightweight Federated Learning and Blockchain System for Industrial IoT Big Data Security

The vast amounts of data produced by industrial IoT devices have recently attracted considerable attention across various scientific disciplines. While big data provides many benefits and applications, it also introduces several challenges that must be resolved to enhance service quality. Artificial intelligence (AI) and blockchain are recognized for their significant potential in strengthening big data security. This study addresses the security of lightweight devices that store large volumes of data but have limited memory and processing power. A lightweight framework for detecting zero-day attacks on resource-constrained yet data-intensive industrial devices was implemented. The performance of the proposed model was assessed using various metrics: Accuracy, Precision, Recall, F1-score, Evaluation time, Gas cost, Loss function, Transaction cost, and its performance compared to the state-of-the-art models. Federated learning (FL) simulation was conducted over 20 rounds, resulting in an overall global model learning accuracy of 97.7%, an F1-score of 97.1 %, a precision of 97.1 %, and a recall value of 97.7%, with a total evaluation time of 899.3 minutes. Additionally, all global model trainable parameters were efficiently deployed over the blockchain public network on the Ethereum smart contract. A security audit carried out showed an average score of 91 %.