Enhancing IoT security: a collaborative framework integrating federated learning, dense neural networks, and blockchain

The recent expansion of the IoT ecosystem has not only significantly increased connectivity but also introduced new security challenges. To address emerging security challenges, this study proposes a framework that merges the decentralized methodologies of federated learning (FL) and Blockchain. The framework is rigorously tested and validated on the N-BaIoT Dataset employing dense neural networks (DNNs) and logistic regression (LR). This approach decentralizes the training of machine learning (ML) models by distributing the process across individual IoT devices, this enhances the security and privacy of data. The use of Blockchain ensures transparent and secure management of these decentralized models, adding an extra layer of protection against tampering. In addition, this research introduces two novel metrics, namely the Security Efficacy Metric and the Comparative Improvement Factor, which provide a quantitative foundation for evaluating the performance of the proposed framework. The examination of the proposed framework through LR and DNNs demonstrates significant results. The LR model achieved a global accuracy of 99.98%, with an average client data size of 440.95 MB and a model size of 0.00088 MB. Meanwhile, the DNN model exhibited a global accuracy of 99.99%, with an average client data size of 551.95 MB and a model size of 0.09 MB. This research contributes to IoT security by integrating LR and DNNs within the FL setup, complemented by blockchain technology, signifying a substantial advancement in the dynamic IoT ecosystem.