Federated learning (FL) is a distributed machine learning approach where each participating node is referred to as a client. Clients are expected to use only local data to train their local machine-learning models. Nevertheless, a client’s poor participation can have a significant impact on FL quality. Therefore, the clients should be carefully chosen, and the selection process should be dynamic and efficient. The majority of proposed methods depend on centrally gathering resource information from nodes, and then selecting clients accordingly. To overcome the drawbacks of the centralization selection paradigm, we propose a decentralized approach in which neighboring nodes cooperate under the guidance of a leader to gather data about a node’s resources. Further, the leader uses a lightweight deep learning model on the collected data to select clients. Compared to the conventional client selection method, our approach speeds up the convergence of the model and consumes less energy by minimizing the number of rounds.