Strategic Intelligent Client Selection in Federated Learning: Optimal Balance Between Communication Efficiency and Model Accuracy

Federated Learning (FL) faces significant communication bottlenecks when deployed across resource-constrained edge devices. This study introduces an intelligent client selection framework that dynamically identifies and selects the most valuable clients for global model aggregation through a metaheuristic-driven optimization process. The proposed method employs a multi-criteria quality assessment mechanism to evaluate client contributions based on accuracy, loss, and historical improvement, and selectively aggregates updates from top-performing participants. Experimental results on the CIFAR-۱۰ dataset under non-Independent and Identically Distributed (non-IID) conditions demonstrate that the approach reduces communication costs by ۸۰% and energy consumption by ۷۵%, while maintaining ۷۵.۲۷% model accuracy, only ۱.۷۳% lower than full client participation. Overall, the method achieves an exceptional trade-off ratio of ۶.۸, significantly outperforming conventional random selection strategies and providing a practical foundation for metaheuristic-based optimization in FL.

کلیدواژه ها:

Federated Learning ، Client Communication Selection ، Efficiency ، Edge Computing ، Resource Optimization ، Top-K Selection