Performance Analysis of Federated Learning in wireless networks

With the proliferation of connected devices and the increase in the use of Machine Learning(ML), more confidential data is being generated. Traditional ML, whereby data is sent to a server for training and processing using models, is becoming less suitable due to privacy concerns. Thus, distributed approaches such as Federated Learning(FL) are becoming more popular. In the latter approach, the model is sent to the clients, where it is trained using the client's data. The updated model is sent to a server to be aggregated. FL is expected to be used extensively in wireless networks. Therefore, researchers are interested in optimizing Federated Learning for wireless networks. This paper aims to study the performance of FL in terms of accuracy and amount of data exchanged in a wireless network considering the impact of delay using different datasets. The accuracy of the FL model was found to be reliable when benchmarked to the centralized approach(less than 0.1 difference in accuracy). The data transfer size with FL was also significantly smaller than in the centralized approach for all the tested datasets.