PSO-Driven Client Selection for Federated Learning in IDS applications

Federated Learning has emerged as a highly promising distributed and collaborative learning paradigm, enabling local clients to train models without sharing their raw data. However, the global model’s performance is often impacted by heterogeneity and variability at the client level, making the selection of clients in each training round a critical factor for overall model effectiveness.In this paper, we propose a novel client selection strategy within the federated learning framework that leverages Particle Swarm Optimization (PSO) to identify the most suitable participants for training a high-performing model in fewer communication rounds. The PSO-based algorithm dynamically adjusts the contribution weights of client models based on their performance and stability metrics, aiming to improve global model accuracy, accelerate convergence, and enable smarter, adaptive model aggregation.We apply this approach specifically to develop a collaborative intrusion detection system for cybersecurity in Edge IIoT environments. Clients are selected based on key criteria such as data quality and individual model performance. Experimental evaluation on a real-world Edge IIoT dataset demonstrates that our solution achieves over 91% accuracy in the global model while reducing the number of communication rounds by approximately 40% compared to the traditional Federated Averaging (FedAvg) method.These results highlight that the proposed approach not only significantly boosts the accuracy of the global model but also accelerates its convergence, making it a robust and efficient solution for Intrusion Detection Solutions (IDS) applications.