Enhancing Cloud Security with Graph Neural Network-Driven Anomaly Detection in Multi-Tenant Environments

Abstract

This paper addresses the key problem of anomaly detection in multitenant cloud environments and proposes an intelligent detection method based on graph neural networks. The study first analyzes the complexity and security risks faced during the process of resource sharing among multiple tenants. It points out that traditional methods based on rules or single features fail to accurately capture tenant interactions and potential abnormal patterns. On this basis, the multi-tenant environment is abstracted into a graph structure, where nodes represent tenants or resources and edges represent interactions or dependencies. Graph neural networks are then used to take advantage of information propagation and feature aggregation, enabling global modeling and representation of tenant behaviors. The method combines graph convolution and attention mechanisms, allowing the model to dynamically focus on key neighbor nodes within multi-level dependencies. This enhances sensitivity and discriminative ability toward abnormal behaviors. At the same time, a joint loss function is constructed to balance feature reconstruction and regularization constraints. This design effectively strengthens the generalization ability and robustness of the model. To verify the effectiveness of the proposed method, systematic experiments were conducted on public cloud computing datasets. Sensitivity analyses were also performed on hyperparameters and environmental conditions. The results show that the method outperforms baseline models in key metrics such as AUC, ACC, F1-Score, and Precision. It maintains stable performance under different conditions. These findings confirm that the proposed anomaly detection framework with graph neural networks has strong practical value and application potential in complex multi-tenant cloud scenarios.