Privacy preserving personalized blockchain reliability prediction via federated learning in IoT environments

摘要

The integration of blockchain and the Internet of Things (IoT) is seen as having significant potential. In IoT Environments, Blockchain builds a trusted environment for IoT information sharing, where information is immutable and reliable. In particular, when edge devices are connected to a blockchain network, they need to be connected to reliable blockchain peers for synchronizing with valid data. Therefore, blockchain reliability prediction has gained attention owing to its ability to help users find highly reliable blockchain peers. Contextual information has been considered useful in many studies for generating highly personalized blockchain reliability predictions. However, these contextual factors are privacy-sensitive, and therefore disclosing them to third parties is risky. To address this challenge, we propose a privacy-preserving personalized blockchain reliability prediction model through federated learning neural collaborative filtering (FNCF) in IoT. Our model allows users to achieve user privacy protection without passing data to a third party and provides personalized predictions for users. We can also leverage the power of edge computing to enable a fast data processing capability and low latency required by IoT applications. Finally, our model was evaluated using a set of experiments based on real-world datasets. The experimental results show that the proposed model achieves high accuracy, efficiency, and scalability.