Fuzzy-Based Client Selection for Enhanced Model Convergence in Federated Learning

Federated learning enables the collaborative training of a global model while allowing the data of multiple clients to remain local. This characteristic renders federated learning highly sensitive to the quality of client data, computational performance, and network conditions. In real-world scenarios, there can be significant disparities in data quality, computational capabilities, and network conditions among clients. Such heterogeneity can severely impact the convergence speed and quality of the global model. To address these challenges, we propose a client selection algorithm that comprehensively considers various metrics of clients. Utilizing fuzzy reasoning, this algorithm analyzes and evaluates clients, enabling the server to select those most conducive to federated training. Extensive experiments conducted across multiple datasets demonstrate that, compared to other client selection algorithms, our algorithm not only marginally improves the accuracy of the global model but also significantly enhances its convergence speed.