This paper introduces a new protocol called FedCS (Federated Learning with Client Selection) for Federated Learning (FL) in a mobile edge computing (MEC) framework. The goal is to efficiently train high-performance machine learning (ML) models while preserving client privacy, especially in practical cellular networks with heterogeneous clients. FedCS addresses the inefficiencies caused by clients with limited computational resources or poor wireless channel conditions by actively managing client resources and optimizing the client selection process. The protocol sets deadlines for clients to download, update, and upload models, allowing the server to aggregate as many client updates as possible within these deadlines. The authors evaluate FedCS using large-scale image datasets and simulate MEC environments, demonstrating that it significantly reduces the training time compared to the original FL protocol. The experimental results show that FedCS achieves higher classification accuracies and completes training faster, even under non-iid data distribution. The study highlights the importance of dynamic resource management and client selection in improving the efficiency of FL in resource-constrained environments.This paper introduces a new protocol called FedCS (Federated Learning with Client Selection) for Federated Learning (FL) in a mobile edge computing (MEC) framework. The goal is to efficiently train high-performance machine learning (ML) models while preserving client privacy, especially in practical cellular networks with heterogeneous clients. FedCS addresses the inefficiencies caused by clients with limited computational resources or poor wireless channel conditions by actively managing client resources and optimizing the client selection process. The protocol sets deadlines for clients to download, update, and upload models, allowing the server to aggregate as many client updates as possible within these deadlines. The authors evaluate FedCS using large-scale image datasets and simulate MEC environments, demonstrating that it significantly reduces the training time compared to the original FL protocol. The experimental results show that FedCS achieves higher classification accuracies and completes training faster, even under non-iid data distribution. The study highlights the importance of dynamic resource management and client selection in improving the efficiency of FL in resource-constrained environments.