SSHAD: Self-Supervised Hybrid Model for Anomaly Detection in Graphs

Graph anomaly detection (GAD) is essential for detecting anomalies in complex networks, with applications in fraud detection, cybersecurity, and social network analysis. However, existing approaches often struggle with limited feature extraction, reliance on labeled data, and scalability when applied to large graphs. Additionally, they fail to address both structural and contextual anomalies comprehensively. To overcome these challenges, we propose SSHAD, a self-supervised hybrid model that integrates GraphSAGE for scalable neighborhood aggregation, GAT for attention-based critical relationship modeling, and VGAE for robust latent graph reconstruction. SSHAD’s self-supervised framework eliminates the need for labeled data, enabling it to adapt to diverse graph domains. It effectively detects node-level and edge-level anomalies, offering a unified solution to structural and contextual irregularities. Extensive evaluations on benchmark datasets (ACM and Cora) highlight SSHAD’s exceptional performance, achieving AUC scores of 97.24% and 90.22%, surpassing state-of-the-art methods in graph anomaly detection. We also provide a comprehensive ablation study to evaluate the contribution of each model component. These results establish SSHAD as a scalable, robust, and versatile solution for real-world graph anomaly detection tasks.