Abstract: Federated Learning (FL) has emerged as a promising paradigm for training machine learning models collaboratively while preserving user data privacy. It enables decentralized learning across multiple clients without sharing raw data. However, challenges such as heterogeneous data distributions and varying client capabilities can hinder model accuracy, convergence speed, and communication efficiency. This paper investigates intelligent client sampling techniques that consider factors like network conditions and model quality to enhance training performance. It reviews and compares random, deterministic, and adaptive sampling methods, highlighting their trade-offs. Emphasis is placed on FL's application in healthcare informatics, where client selection remains underexplored despite its critical importance. Thus, the proposed Class Topper Optimization (CTO) approach is the foundation of the novel federated learning strategy presented in the suggested methodology. This paper utilizes medical data on multiple clients as inputs for the projected framework. In this suggested technique, the ANNs and CNNs operate as localized models for training homogenous customers. After that, the global model is updated frequently by receiving updates from locally trained models. The global model is implemented by employing LSTM networks. The precision level is around 92% and the accuracy rate is roughly 93%. Hence, in federated learning, the ideal client selection method is the optimization-based method.

Key words: federated learning (FL), client selection, system heterogeneity, data heterogeneity, optimization