Adaptive Federated Learning with Privacy Preservation for Robust Anomaly Detection in Multi-Cloud Environments

Multi-cloud deployments face significant security challenges due to fragmented visibility and regulatory constraints on data sharing.This paper proposes a novel Federated Learning (FL) framework for privacy-preserving anomaly detection across heterogeneous cloud environments.The proposed approach combines adaptive federated aggregation (AFA) with a hybrid CNN-LSTM model, differential privacy, and homomorphic encryption to address non-IID data distributions, communication overhead, and privacy risks.Evaluations using synthesized AWS, Azure, and GCP workload traces demonstrate 92.3% F1-score (13.7% improvement over FedAvg) while reducing communication overhead by 63% and resisting model inversion attacks with =1.0 differential privacy.The framework maintains compliance with GDPR/HIPAA by design, eliminating raw data transmission.Comparative analysis reveals 28% faster convergence than centralized approaches in asymmetric network conditions, establishing FL as a viable paradigm for cross-cloud security analytics.