Performance Evaluation of Hierarchical Federated Learning on Resource-Constrained IoT Systems

The Internet of Things (IoT) has transformed how smart cities operate, significantly improving their inhabitants' efficiency and overall quality of life. However, the massive volume of sensitive data generated by IoT devices presents challenges, including privacy concerns and communication overheads. Traditional centralized data processing can compromise privacy and require significant communication, leading to scalability problems and power drains. Federated Learning (FL) offers a solution by processing data locally and transmitting only the model parameters. Vanilla FL faces challenges in IoT environments due to latency, bandwidth constraints, and power drain. Hierarchical FL (HFL) effectively addresses these issues by hierarchically leveraging the processing capabilities of both cloud and edge servers to optimize resource utilization and efficiently minimize latency. This paper evaluates HFL using IoT-derived datasets, develops and implements the HFL framework for resource-constrained IoT systems, and conducts the first known HFL tests on relevant datasets to demonstrate its performance.