Dynamic Aggregation and Personalized Federated Learning for Heterogeneous Threat Detection in Privacy-Preserving Networks

As modern networks face increasingly sophisticated and heterogeneous cyber threats, ensuring effective and privacy-preserving threat detection remains a critical challenge. detection remains a critical challenge. To address these challenges, this paper proposes a novel framework that integrates Personalized Federated Learning (PFL) with a Dynamic Aggregation Mechanism (DAM). Learning (PFL) with a Dynamic Aggregation Mechanism (DAM). The PFL component enables each client to fine-tune the global model based on its local threat landscape, ensuring more effective detection of region- specific threats. Meanwhile, DAM dynamically adjusts the aggregation weights of clients during global model updates, based on the quality and relevance of each client's model. Meanwhile, DAM dynamically adjusts the aggregation weights of clients during global model updates, based on the quality and relevance of each client's data. This approach ensures that high-quality data from certain clients have a greater influence on the global model, while minimizing the impact of low-quality or noisy data. The proposed framework is evaluated in the context of heterogeneous network threat detection using simulated and real-world datasets. detection accuracy and adaptability to diverse client environments, while maintaining stringent privacy guarantees through local data preservation.