<p>Unsupervised anomaly detection in static attributed networks is a crucial research area in network science, with applications spanning cybersecurity, social network analysis, and beyond. However, the complex structures and rich node attributes of these networks pose significant challenges for traditional methods, which often struggle to capture multifaceted relationships and detect both structural and contextual anomalies. To address these challenges, we introduce the Dual Variational Autoencoder Graph Neural Network (DVAE-GNN), a novel framework for unsupervised anomaly detection in static attributed networks leveraging the strengths of variational autoencoders (VAEs) and graph neural networks (GNNs) to effectively capture both attribute and structural information. The DVAE-GNN framework employs a structure autoencoder with graph neural network to learn structural representations, while the attribute variational autoencoder models the probabilistic distributions of node attributes to learn node representations. We also incorporate a multi-head attention mechanism to enable the model to prioritize and aggregate relevant information from a node’s neighbors, enhancing its ability to identify structural anomalies. DVAE-GNN is trained end-to-end using a joint loss function that balances network structure reconstruction and attribute reconstruction to learn a unified representation that encapsulates the complex interactions between node attributes and the network structure, thus enabling effective detection of both structural and attribute-based anomalies with heightened sensitivity and accuracy. Our extensive experiments on multiple real-world and synthetic datasets reveal that DVAE-GNN consistently achieves high average performance in ROC-AUC scores, showcasing its robustness and adaptability to diverse network characteristics. The source code for DVAE-GNN is openly available at: https://github.com/hpiridi/dvaegnn.</p>