Federated Learning Methodology for Indoor Location Systems in Multi-Layer Architecture

Recent research has shown that federated learning offers numerous benefits, including improved privacy, reduced communication costs, and enhanced scalability. Our proposed architecture and methodology for indoor Bluetooth localization systems leverages these advantages by utilizing federated learning in the fog layer, leading to increased accuracy in classifier models. To do so, we propose an architecture for indoor Bluetooth localization systems based on edge, fog, and cloud computing layers with temporal windows, utilizing federated learning in the fog layer of the architecture. Additionally, we present a methodology in which data is collected and aggregated on the edge layer, the localization model is trained and deployed on fog nodes, and a conflict resolution method is proposed in the cloud. A localisation case study is evaluated to determine the optimal parameters of the methodology, and to check the consistency and speed of a model generated with the methodology. According to the experiment performed, increasing the time window for data aggregation in both the edge and fog layers results in higher classifier accuracy.