A trustworthy BCFL in indoor localization system

Abstract

To achieve an exact indoor localization effect, the fingerprinting-based machine learning method is a promising choice. But in the process of training arise some privacy and security concerns. To handle the privacy concern, the common choice is adopting a federated learning (FL) framework. However traditional FL frameworks are helpless against remaining security concerns, such as malicious attacks and single-point failure.

To solve these concerns, we design a trustworthy Blockchain-based federated learning (BCFL) framework called DFLoc. Specifically, the central server in traditional federated learning is replaced with a PoS blockchain, to solve the single-point failure. Malicious attacks are figured out by the elaborated DFLoc Validator Mechanism.

To evaluate the performance of our proposed framework in detail, we conduct extensive experiments using a real-world dataset of fingerprinting-based indoor localization called UJIIndoorLoc. The experiment results demonstrate that our DFLoc can effectively mitigate the challenges of malicious attacks and singlepoint failure in a 3D environment when compared with the traditional centralized FL systems.