Privacy-Preserving Cyber Threat Intelligence Correlation across Distributed Cloud Enterprise Networks Using Federated Machine Learning

Abstract

Distributed cloud enterprise networks generate large volumes of cybersecurity telemetry from cloud workloads, identity services, network devices, applications, and security monitoring platforms. Correlating these heterogeneous data sources is essential for identifying coordinated attacks, detecting emerging threats, and improving enterprise security operations. However, conventional centralized threat intelligence approaches require organizations to share sensitive operational data, creating privacy, regulatory, and confidentiality challenges. This research proposes a privacy-preserving cyber threat intelligence correlation framework based on federated machine learning, enabling multiple enterprises to collaboratively identify cyber threats without directly exchanging their raw security data. The proposed methodology integrates local feature extraction, federated model training, secure model aggregation, privacy-enhancing mechanisms, and cross-enterprise threat correlation. Each participating organization trains a local detection model using its own telemetry, while encrypted model updates are aggregated to improve collective threat detection capabilities. Differential privacy and secure aggregation are incorporated to reduce the risk of exposing sensitive information through exchanged updates. The framework evaluates detection accuracy, precision, recall, F1-score, communication overhead, model convergence, and privacy protection under heterogeneous network conditions. Additional assessments examine its resilience against poisoning attacks, non-independent data distributions, and unreliable participants. The proposed approach aims to improve collaborative cyber defense while preserving organizational data sovereignty. Its effectiveness must be established through reproducible experiments comparing federated learning with centralized and independent local-learning baselines.